Blame for SignalDoc.csv

Acronym,Cat.Signal,Predictability in OP,Signal Rep Quality,Authors,Year,LongDescription,Journal,Cat.Form,Cat.Data,Cat.Economic,SampleStartYear,SampleEndYear,Acronym2,Evidence Summary,Key Table in OP,Test in OP,Sign,Return,T-Stat,Stock Weight,LS Quantile,Quantile Filter,Portfolio Period,Start Month,Filter,Notes,Detailed Definition,GScholarCites202509
AbnormalAccruals,Predictor,1_clear,2_fair,Xie,2001,Abnormal Accruals,AR,continuous,Accounting,accruals,1971,1992,AccrAbn,t=8 port sort w/ nonstandard data lag,3,port sort size adjusted nonstandard,-1.0,0.916666667,8.43,EW,0.1,,12.0,6.0,,OP is aggressive and lags accounting data by only 3 months (p361) instead of the usual 6.  This likely accounts for our relative underperformance.,"Define Accruals as net income (ib) minus operating cash flow (oancf), divided by average total assets (at) for years t-1 and t. If oancf is missing, replace operating cash flow with funds from operations (fopt) minus the annual change in total current assets (act) plus the annual change in cash and short-term investments (che) plus the annual change in current liabilities (lct) minus the annual change in debt in current liabilities (dlc). For each year t and 2-digit sic code, regress Accruals on: the inverse of average total assets for year t-1, the change in revenue (sale) from year t-1 to t divided by total assets for t-1, propery plant and equipment (ppegt) divided by total assets for t-1.  AbnormalAccrual is the residual from this cross-sectional regression.  See code for more details.",139.0
Accruals,Predictor,1_clear,1_good,Sloan,1996,Accruals,AR,continuous,Accounting,accruals,1962,1991,Accruals,t > 4 in port sort CAPM alpha 12 month holding,6 year t+1,port sort size adjusted,-1.0,0.866666667,4.71,EW,0.1,,12.0,6.0,abs(prc)>5,Table 6 year t+1 hedge.  Only size adjusted and CAPM adjusted.,Annual change in current total assets (act) minus annual change in cash and short-term investements (che) minus annual change in current liabilities (lct) minus annual change in debt in current liabilities (dlc) minus change in income taxes (txp). All divided by average total assets (at) over this year and last year. Exclude if abs(prc) < 5.,56.0
AccrualsBM,Predictor,1_clear,1_good,Bartov and Kim,2004,Book-to-market and accruals,RFQA,discrete,Accounting,valuation,1980,1998,AccrualsBM,t=5.5 in long-short,3 mean diff 1-2,LS port,1.0,0.206,5.5,EW,0.2,,12.0,6.0,,,"Binary variable equal to 1 if stock is in the highest Accrual quintile and the lowest BM quintile, and equal to 0 if stock is in the lowest Accrual quintile and the highest BM quintile. Exclude if book equity (ceq) is negative.",22.0
Activism1,Predictor,1_clear,1_good,Cremers and Nair,2005,Takeover vulnerability,JF,continuous,13F,other,1990,2001,Activism1,t=3.1 in port sort,3A VW BLOCK=4,port sort CAPM alpha,1.0,0.9025,3.13,VW,0.25,,1.0,6.0,,works a bit better EW in Tab 3,"24 minus Governance Index (G). Set to missing if G is missing, or if not in the highest quartile of institutional ownership (maxinstown\_perc), or if dual share class.",2036.0
AM,Predictor,1_clear,1_good,Fama and French,1992,Total assets to market,JF,continuous,Accounting,valuation,1963,1990,AM,t=5.7 in univar reg,3 Ln(A/ME),univariate reg,1.0,,5.69,EW,,,12.0,6.0,,,Total assets (at) divided by market value of equity.,29625.0
AnalystRevision,Predictor,1_clear,2_fair,"Hawkins, Chamberlin, Daniel",1984,EPS forecast revision,FAJ,continuous,Analyst,earnings forecast,1975,1980,AnalystRevision,t=3.2 in long only CAPM alpha,10 mean,long only port CAPM alpha,1.0,0.460583333,3.169151376,EW,0.2,,1.0,12.0,,OP only longs top 20 stocks according to signal.  We were more flexible.  Sample is very short but the results seem robust.,"keep fpi == ""1"", last obs each month.  Signal is meanest / last month's meanest.",117.0
AnnouncementReturn,Predictor,1_clear,1_good,"Chan, Jegadeesh and Lakonishok",1996,Earnings announcement return,JF,continuous,Price,earnings event,1977,1992,AnnounRet,t=9.3 in regression,7 ABR,mv reg,1.0,,9.25,EW,,,1.0,12.0,,Table 4 has port sort but no t-stats.  Tab 7 has huge t-stats in regressions,Get announcement date for quarterly earnings from IBES (fpi = 6). AnnouncementReturn is the sum of (ret - mktrf + rf) from one day before an earnings announcement to 2 days after the announcement.,3314.0
AssetGrowth,Predictor,1_clear,1_good,"Cooper, Gulen and Schill",2008,Asset growth,JF,continuous,Accounting,investment,1968,2003,InvestAG,t=8.5 in port sort,2B.1 year 1,port sort,-1.0,1.73,8.45,EW,0.1,,12.0,6.0,,"Also works VW (t=5, Tab2.B.2)",Annual growth rate of total assets (at),2226.0
BetaLiquidityPS,Predictor,1_clear,1_good,Pastor and Stambaugh,2003,Pastor-Stambaugh liquidity beta,JPE,continuous,Price,liquidity,1968,1999,BetaLiquidityPS,t=2.54 in VW port sort CAPM alpha,4A CAPM 10-1,port sort CAPM alpha,1.0,0.533333333,2.54,VW,0.1,,12.0,12.0,abs(prc)>5,,"Monthly excess return (ret -rf) regressed on innovations in liquidity from Pastor's website (\url{https://faculty.chicagobooth.edu/lubos.pastor/research/liq_data_1962_2018.txt}). Use 60 month rolling window regression, and require at least 36 non-missing observations.",7459.0
BetaTailRisk,Predictor,1_clear,1_good,Kelly and Jiang,2014,Tail risk beta,RFS,continuous,Price,risk,1963,2010,BetaTailRisk,Tab4A t-stat 2.48,4A EW,port sort,1.0,0.33,2.48,EW,0.2,,12.0,12.0,abs(prc)>5,Also works VW or monthly,"Each month, compute the 5th percentile over daily returns over all firms. For all daily return observations with return below that 5th percentile, compute the average of (log(ret/5th percentile of cross-sectional return distribution). Call that average tailEX. BetaTailRisk is the coefficient of a 120-month rolling regression of a firm's stock return on tailEX. Exclude if price less than 5 or share code greater than 11.",1020.0
betaVIX,Predictor,1_clear,1_good,Ang et al.,2006,Systematic volatility,JF,continuous,Price,volatility,1986,2000,betaVIX,t=3.9 in port sort,1A,port sort,-1.0,1.04,3.9,VW,0.2,,1.0,6.0,,Tab I has port sorts,Coefficient on daily change in the VIX of a 1-month rolling window regression of daily stock excess returns on market return and the daily change in the CBOE S&P 100 volatility index (downloaded from FRED). Require at least 15 non-missing observations.,6323.0
BM,Predictor,1_clear,1_good,Stattman,1980,"Book to market, original (Stattman 1980)",Other,continuous,Accounting,valuation,1962,1976,BM,risk-adjusted portfolio independence test,3,risk-adjusted portfolio independence test,1.0,,,EW,0.1,,12.0,6.0,"exchcd%in%c(1,2)","OP actually forms portfolios at the end of March based on December FYE data and drops stocks with non-Dec FYE.  But for consistency with FF1992, we use June as the port form month (the irony is noted).  OP also drops certain SIC codes, which we should implement at the portfolio stage, in time.  We previously were citing Rosenberg, Reid, Lanstein 1985, but Stattman 1980 should get more credit.  RRL does not describe its methods in much detail too.",Log of tangible book equity (ceqt) over market equity matched at FYE,1555.0
BMdec,Predictor,1_clear,1_good,Fama and French,1992,Book to market using December ME,JF,continuous,Accounting,valuation,1963,1990,BMdec,t=5.71 in univariate reg,3 Ln(BE/ME),univariate reg,1.0,0.5,5.71,EW,,,12.0,6.0,,,BM using most recent December value of market equity.,5940.0
BookLeverage,Predictor,1_clear,1_good,Fama and French,1992,Book leverage (annual),JF,continuous,Accounting,leverage,1963,1990,BookLev,t=5.3 in mv reg,3 Ln(A/BE),mv reg,-1.0,,5.34,EW,,,12.0,6.0,,,"Total assets (at) divided by book value of equity plus deferred taxes (txditc) and preferred stock. Equity is shareholder equity (seq) if available, or book equity (ceq) plus preferred stock (pstk, if missing pstkrv, if missing pstkl), or total assets minus total liabilities (lt).",29625.0
BPEBM,Predictor,1_clear,1_good,"Penman, Richardson and Tuna",2007,Leverage component of BM,JAR,continuous,Accounting,leverage,1963,2001,BMlev,t=4.1 in univariate reg,1D,univariate reg,-1.0,,4.1,EW,,,12.0,6.0,abs(prc)>5,B/P-NOA/P^NOA in OP.  Table 1D shows strong port sort.  Tab3A shows t-stat = 4.1 in uni reg. OP drops extreme obs but we don't.,"BP - EBM, where BP = (ceq + tstkp - dvpa)/(shrout*abs(prc)), and EBM is defined above. Exclude if price less than 5.",550.0
Cash,Predictor,1_clear,1_good,Palazzo,2012,Cash to assets,JFE,continuous,Accounting,asset composition,1972,2009,Cash,t=2.14 in port sort but strong with adjustments,4,port sort,1.0,0.69,2.14,EW,0.1,,1.0,6.0,,"Table 4 has long-short returns with t-stat of 2.14 for raw EW. Much stronger after factor adjustments. Other tables show double sorts, alphas, Table 4, Panel A, column DeltaCH",Ratio of quarterly cash and short-term investments (cheq) and total assets (atq).,500.0
CashProd,Predictor,1_clear,1_good,Chandrashekar and Rao,2009,Cash Productivity,WP,continuous,Accounting,profitability alt,1963,2003,CashProd,t=3.6 in regression,4A \eta,mv reg,-1.0,,3.6,EW,,,1.0,6.0,,Stats are from WP version.,Calculate market value of equity (mve_c) as absolute price (prc) times number of shares outstanding (shrout). Cash productivity is equal to the difference between mve_c and total assets (at) divided by cash and short-term investments (che).,2.0
CBOperProf,Predictor,1_clear,1_good,Ball et al.,2016,Cash-based operating profitability,JFE,continuous,Accounting,profitability,1963,2014,ProfCash,t=3.2 in port sort,4A Cash-based 10-1,port sort,1.0,0.47,3.17,VW,0.1,NYSE,12.0,6.0,,This is operating prof with working cap and R&D adjustments.,"Revenue (revt) minus cost (cogs) - (administrative expenses (xsga) - R&D expenses (xrd))  minus annual change in receivables (rect), annual change in investment (invt) and annual change in prepaid expenses, plus annual change in current deferred revenue (drc), long-term deferred revenue (drlt), accounts payable (ap) and accrued expenses (xacc), all divided by total assets (at) in year t. Replace all variables in the numerator with 0 if they are missing. Exclude if share code is greater 11, market value of equity, BM or total assets are missing, or if SIC code between 6000 and 6999.",533.0
CF,Predictor,1_clear,1_good,"Lakonishok, Shleifer, Vishny",1994,Cash flow to market,JF,continuous,Accounting,valuation,1968,1990,CF2Price,t=3.4 in port sort,6 panel 1,port sort,1.0,0.658333333,3.379,EW,0.1,,12.0,6.0,"exchcd%in%c(1,2)",,Net income (ib) plus depreciation (dp) divided by market equity. Exclude NASDAQ stocks.,7650.0
cfp,Predictor,1_clear,1_good,"Desai, Rajgopal, Venkatachalam",2004,Operating Cash flows to price,AR,continuous,Accounting,valuation,1973,1997,CFOper2Price,t=2.77 in port sort,2E R1,port sort,1.0,1.275,2.77,EW,0.2,,12.0,6.0,,,"Operating cash-flow (oancf) divided by market value of equity. If operating cash-flow is missing, replace by difference betwee net income (ib) and level of accruals, where the latter is the annual change in current assets (act) minus the annual change in cash and short-term investments (che), minus the annual change in current liabilities (lct) plus the annual change in debt in current liabilities (dlc) plus the annual change in payable income taxes (txp) plus depreciation (dp).",649.0
ChangeInRecommendation,Predictor,1_clear,4_lack_data,Jegadeesh et al.,2004,Change in recommendation,JF,continuous,Analyst,recommendation,1985,1998,ChRecomm,"p<0.01 in LS port, but we lack the data",3C,LS nonstandard data,1.0,0.225,,EW,0.2,,1.0,12.0,,"OP sample is 1985-1998 using Zack's, but our IBES recommendations only begins in 1993.  OP is binary, but we follow MP.  Even though sample is super short for us, it seems to work, and is even mostly monotonic.","keep last ireccd each month, then average across analysts for each firm-month.  Define opscore as 6-ireccd.  Signal is opscore - last month's opscore.",1466.0
ChAssetTurnover,Predictor,1_clear,1_good,Soliman,2008,Change in Asset Turnover,AR,continuous,Accounting,sales growth,1984,2002,ATurnGr,t=5 in mv reg,7 Model 1 DeltaATO,mv reg,1.0,,5.12,EW,,,12.0,6.0,abs(prc)>5,"Tab 7 Delta ATO, t-stat 2.5 with controls, no sorts.",Annual change in AssetTurnover (defined above). Exclude if price less than 5.,919.0
ChEQ,Predictor,1_clear,1_good,Lockwood and Prombutr,2010,Growth in book equity,JFR,continuous,Accounting,investment,1964,2007,BEgrowth,t=5.38 in EW port sort,4A SUSG,port sort,-1.0,0.8,5.38,EW,0.2,,12.0,6.0,,"Table 4, panel A, take difference between High and low SUSG returns, tstat is reported. Paper also offers alphas and FM regressions.",Ratio of book equity (ceq) to book equity in the previous year. Include only if book equity is positive this year and last year.,111.0
ChForecastAccrual,Predictor,1_clear,1_good,Barth and Hutton,2004,Change in Forecast and Accrual,RAS,discrete,Analyst,earnings forecast,1981,1996,ChFAccrual,p-val < 0.001 in port sort,3B consistent: also 3A,port sort,1.0,2.375,,EW,,,12.0,6.0,,"Closest is Tab 3B.  OP basically does a double sort on accruals and revisions. p-val < 0.001, but no t-stat.  Our t-stat is enormous, suggesting the p-value is much less than 0.001.","Within upper half of Accruals distribution, equal to 1 if mean earnings estimate increased relative to the previous month. 0 if it decreased.",394.0
ChInv,Predictor,1_clear,1_good,Thomas and Zhang,2002,Inventory Growth,RAS,continuous,Accounting,investment alt,1970,1997,Invntory,t>2.6 in port sort,1 Delta Invent,port sort size adjusted no LS,-1.0,0.949166667,,EW,0.1,,12.0,6.0,,Table 1 \Delta Invent row has decile hedge size adjusted returns. T-stats are missing though. Main results use panel regression coefficient and stars.,12 month change in inventory (invt) divided by average total assets.,859.0
ChInvIA,Predictor,1_clear,1_good,Abarbanell and Bushee,1998,Change in capital inv (ind adj),AR,continuous,Accounting,investment growth,1974,1988,InvestGr,t=2.9 in mv reg,2b RCAPX,reg,-1.0,,2.914,EW,,,12.0,6.0,,,"Growth in capital expenditure (capx) minus average growth in capital expenditure in the same industry (two-digit SIC). If capx is missing, capital expenditure is defined as the annual change in property, plant and equipment (ppent). Capital expenditure growth is defined as the percentage growth of capx today relative to the average capx over the previous two years (.5*(capx$_{t-1}$ + capx$_{t-2}$), or as percentage growth relative to the previous year only if t-2 is missing.",1123.0
ChNAnalyst,Predictor,1_clear,1_good,Scherbina,2008,Decline in Analyst Coverage,ROF,discrete,Analyst,earnings event,1982,2005,ChNAnalyst,t > 3 in port sort FF3 alpha for small stocks,2 alphas return diff,port sort FF3 alpha,-1.0,0.46,3.34,VW,,,1.0,12.0,abs(prc) > 5,"Tab 2 says t=0.3 for all stocks, but t>3 for size quintiles 1-2","Using IBES forecasts, keep if fpi = 1, fpedats not missing, and fpedats > statpers + 30.  Binary variable equal to 1 if the number of analysts (numest) for next quarter's EPS estimate decreased relative to three months ago, and 0 if it increased.  Keep if in bottom two size quintiles among all firms to match OP's table.",116.0
ChNNCOA,Predictor,1_clear,1_good,Soliman,2008,Change in Net Noncurrent Op Assets,AR,continuous,Accounting,investment alt,1984,2002,ChNNCOA,t=4.3 in mv reg,7 DeltaNCO,mv reg,-1.0,,5.26,EW,,,12.0,6.0,,"Tab 7 Delta NCO, highly significant with controls,",Twelve-month change in noncurrent operating assets. Noncurrent operating assets is ( (at - act - ivao) - (lt - dlc - dltt) )/at.,919.0
ChNWC,Predictor,1_clear,1_good,Soliman,2008,Change in Net Working Capital,AR,continuous,Accounting,investment alt,1984,2002,NWCgr,t=4.6 in mv reg,7 Model 2 DeltaWC,mv reg,-1.0,,4.61,EW,,,12.0,6.0,abs(prc)>5,"Very strong in multivariate regression (t-stat of above 4.6 in both specifications), no sorts. Main results (Table 7) use annual fama macbeth.",Twelve-month change in net working capital. Net working capital is ( (act - che) - (lct - dlc) )/at,919.0
ChTax,Predictor,1_clear,1_good,Thomas and Zhang,2011,Change in Taxes,JAR,continuous,Accounting,other,1977,2006,TaxGr,t = 11.26 in decile sort,"Table 2, column (1).",port sort,1.0,1.3,11.26,EW,0.1,,3.0,6.0,,,"4-quarter change in quarterly total taxes (txtq), scaled by lagged total assets (at).",196.0
CitationsRD,Predictor,1_clear,2_fair,"Hirschleifer, Hsu and Li",2013,Citations to RD expenses,JFE,discrete,Other,profitability alt,1982,2008,CitationsRD,t=2.6 in FF3 style long-short,9A EMI2,LS FF3 style,1.0,0.26,2.6,VW,,,12.0,6.0,,"Table 9 does ff3 style VW to adjust for size.  Table 8 shows that predictability weak in large firms, so we just focus on small firms and VW to keep this spreadsheet manageable.","Drop obs before 1975.  Set expenses for research and development (xrd) and number of citations (ncit)  to 0 if missing. Calculate Citations to RD as sum of citations over previous 5 years divided by sum of xrd over years $t-3, \ldots, \t-7$.  Drop if denominator = 0,  2 years or less in Compustat, sic in 6000s, ceq < 0.  Double independent sort using (a) size NYSE median (b) terciles of Citations to RD.  Long if in small size, highest tercile of Citations to RD, short if in small size, smallest tercile.",1239.0
CompEquIss,Predictor,1_clear,1_good,Daniel and Titman,2006,Composite equity issuance,JF,continuous,Accounting,external financing,1968,2003,CompEquIss,t=4.4 in mv reg,"3 \iota(t-5,t)",univariate reg,-1.0,,4.39,EW,,,1.0,6.0,abs(prc)>5,,5 year growth rate of market value of equity minus 5 year stock return.,1640.0
CompositeDebtIssuance,Predictor,1_clear,1_good,"Lyandres, Sun and Zhang",2008,Composite debt issuance,RFS,continuous,Accounting,external financing,1970,2005,DebtFinC,t=8.59 in port sort CAPM alpha,5B,port sort CAPM alpha,-1.0,0.523,8.59,EW,0.333333333,,12.0,6.0,,"Table 5B. Uses decile sorts, but then longs top 3 and shorts bottom 3 portfolios.  Table 3 offers hedge return alphas for IPOs and debt issue indicators too.",Log of long-term debt (dltt) plus debt in current liabilties (dlc) minus log of the same variable 5 years ago.,663.0
ConvDebt,Predictor,1_clear,1_good,Valta,2016,Convertible debt indicator,JFQA,discrete,Accounting,external financing,1985,2012,ConvDebt,t > 2.6 in mv reg,4 DCONV,mv reg,-1.0,,4.5,EW,,,1.0,6.0,,"Table 4, DCONV has FM regression results for dummy.  Dummy has opposite sign of covertible proportion, both are significant.  Table 5C has port sort for proprotion, and returns are nonmonotonic and LS t=1.7.  We focus on the dummy but it's a judgement call.",Binary variable equal to 1 if deferred charges (dc) greater than 0 or common shares reserved for convertible debt (cshrc) greater than 0.,124.0
CoskewACX,Predictor,1_clear,1_good,"Ang, Chen and Xing",2006,Coskewness using daily returns,RFS,continuous,Price,risk,1963,2001,CoskewACX,t=2.8 in port sort,8B,port sort,-1.0,0.28,2.76,EW,0.2,,1.0,12.0,"shrcd<=11, exchcd==1","Used primarily as a control variable in the paper, but this paper has details on the signal construction that are not found in Harvey and Siddique.  ACX simply use de-meaned stock returns rather than CAPM residual.  Also uses daily data, and equal weighting.","Signal is the sample counterpart of $E[\tilde{r}_{it} \tilde{r}_{mt}^2]/( SD[\tilde{r}_{it} ] SD[\tilde{r}_{mt}]^2$ where $\tilde{r}_{it}$ is the de-meaned stock return and $\tilde{r}_{mt}$ is the de-meaned market excess return.  Signal is computed using the past year of daily data, and using the NYSE CRSP VW index for the market (dsia), with returns continuously compounded.  See code for details.",1691.0
CPVolSpread,Predictor,1_clear,1_good,Bali and Hovakimian,2009,Call minus Put Vol,MS,continuous,Options,optionrisk,1996,2004,CPVolSpread,t=4 in port sort,Table 3 Panel B,port sort,1.0,1.045,4.2,VW,0.2,,1.0,6.0,"shrcd%in%c(10,11)",OP says NYSE only but we find all stocks gets closer to their numbers,ATM Call vol minus ATM put vol,499.0
dVolCall,Predictor,1_clear,1_good,"An, Ang, Bali, Cakici",2014,Change in call vol,JF,continuous,Options,informed trading,1996,2011,dImpVolCall,t=3.5 in port sort,Table IIA,port sort,1.0,1.09,3.45,EW,0.1,,1.0,12.0,,,"Using OptionM vol surface, 30 day mat, delta = 0.5, find first diff of implied vol",473.0
dVolPut,Predictor,2_likely,1_good,"An, Ang, Bali, Cakici",2014,Change in put vol,JF,continuous,Options,informed trading,1996,2011,dImpVolPut,t=2.0 in port sort,Table IIB,port sort,-1.0,0.42,2.03,EW,0.1,,1.0,12.0,,,"Using OptionM vol surface, 30 day mat, delta = 0.5, find first diff of implied vol",473.0
dCPVolSpread,Predictor,1_clear,1_good,"An, Ang, Bali, Cakici",2014,Change in put vol minus change in call vol,JF,continuous,Options,informed trading,1996,2011,dImpVolSmile,t=7 in port sort,Table IIC,port sort,-1.0,1.68,6.77,EW,0.1,,1.0,12.0,,,Take difference of dVolCall and dVolPut,473.0
CredRatDG,Predictor,1_clear,4_lack_data,Dichev and Piotroski,2001,Credit Rating Downgrade,JF,discrete,Analyst,other,1986,1998,CredRatDG,t=11 in event study w/ special data,4 Downgrade BHAR 3-month,event study 3-month size+BM adjusted nonstandard data,-1.0,1.316666667,11.04,EW,,,1.0,12.0,,"Most of the returns are earned in first 3 days of announcement.  OP uses Moody's Default Risk Service data going back to 1970, but our S\&P Credit Ratings data only goes back to 1978","A downgrade happens if credit rating (splticrm) decreased by at least one notch relative to the previous month.  CredRatDG = 1 if a downgrade happened over the past 3 months.  OP studies Moody's ratings changes between 1970 and 1997, but our data doesn't begin in earnest until 1986, and our sample definitions apply to returns, and this predictor implicitly averages over the past 6 months.",948.0
CustomerMomentum,Predictor,1_clear,1_good,Cohen and Frazzini,2008,Customer momentum,JF,continuous,Other,lead lag,1980,2004,MomCust,t=3.8 in port sort,3A,port sort,1.0,1.578,3.79,VW,0.2,,1.0,12.0,abs(prc)>5,"Tab3A is VW, so we do VW.  3B shows works a bit better EW",Based on firms' principals customers from Compustat Segment data as in Cohen and Frazzini.,1793.0
DelBreadth,Predictor,1_clear,1_good,"Chen, Hong and Stein",2002,Breadth of ownership,JFE,continuous,13F,ownership,1979,1998,DelBreadth,t=4.0 in port sort,4A 1 Quarter,port sort size adjusted,1.0,0.673333333,3.96,EW,0.1,,3.0,6.0,,"Table 4 (chose the cumluative annual return, but they report 1,2,3 and 4 quarters),",Quarterly change in the number of institutional owners (numinstowners) from 13F data. Exclude if in the lowest quintile of stocks by market value of equity (based on NYSE stocks only).,1672.0
DelCOA,Predictor,1_clear,1_good,Richardson et al.,2005,Change in current operating assets,JAE,continuous,Accounting,investment alt,1962,2001,AssetCGr,t=9 in mv reg,8C,univariate reg,-1.0,,8.71,EW,,,12.0,6.0,,"Table 8, but FMB only.  Most results are about accounting rate of return (Table 5-7).","Difference in current operating assets (total current assets (act) minus cash and short-term investments (che)) between years t-1 and t, scaled by average total assets (at) in years t-1 and t.",2296.0
DelCOL,Predictor,1_clear,1_good,Richardson et al.,2005,Change in current operating liabilities,JAE,continuous,Accounting,external financing,1962,2001,LiabCGr,t=4.5 in mv reg,8C,univariate reg,-1.0,,4.49,EW,,,12.0,6.0,,"Table 8, but FMB only.  Most results are about accounting rate of return (Table 5-7).","Difference in current operating liabilities (total current liabilities (lct) minus debt in current liabilities (dlc)) between years t-1 and t, scaled by average total assets (at) in years t-1 and t.",2296.0
DelEqu,Predictor,1_clear,1_good,Richardson et al.,2005,Change in equity to assets,JAE,continuous,Accounting,investment,1963,2001,Eq2AGr,t=6.3 in mv reg,9A,mv reg,-1.0,,6.25,EW,,,12.0,6.0,,Delta Equity in OP.,"Difference in book equity (ceq) between years t-1 and t, scaled by average total assets (at) in years t-1 and t.",2296.0
DelFINL,Predictor,1_clear,1_good,Richardson et al.,2005,Change in financial liabilities,JAE,continuous,Accounting,external financing,1962,2001,FinLiabGr,t=8 in univariate reg,8C,univariate reg,-1.0,,8.01,EW,,,12.0,6.0,,"Table 8, but FMB only.  Most results are about accounting rate of return (Table 5-7).","Difference in financial liabilities (sum of long-term debt (dltt), current liabilitites (dlc) and preferred stock (pstk)) between years t-1 and t, scaled by average total assets (at) in years t-1 and t.",2296.0
DelLTI,Predictor,1_clear,1_good,Richardson et al.,2005,Change in long-term investment,JAE,continuous,Accounting,investment,1962,2001,ChLTI,t=3.4 in mv reg,8C,univariate reg,-1.0,,3.38,EW,,,12.0,6.0,,"Table 8, but FMB only.  Most results are about accounting rate of return (Table 5-7).","Difference in investment and advances (ivao) between years t-1 and t, scaled by average total assets (at) in years t-1 and t.",2296.0
DelNetFin,Predictor,1_clear,1_good,Richardson et al.,2005,Change in net financial assets,JAE,continuous,Accounting,investment alt,1962,2001,DelNetFin,t=6 in unvivariate reg,8B DeltaFin,univariate reg,1.0,,5.85,EW,,,12.0,6.0,,,"Compute the sum of short-term investments (ivst) and investments and advances (ivao) minus the sum of long-term debt (dltt), debt in current liabilities (dlc) and preferred stock capital. Divide the difference between the current and one-year lagged sum by total assets (at) averaged over the current and previous fiscal years.",2296.0
DivInit,Predictor,1_clear,2_fair,"Michaely, Thaler and Womack",1995,Dividend Initiation,JF,discrete,Event,payout indicator,1964,1988,DivInit,t=3.4 in event study,"3 init, to Day 254",event study 12 months,1.0,0.625,3.37,EW,,,1.0,12.0,shrcd <= 11,"We deviate from OP in not imposing an NYSE/AMEX requirement.  This allows our portfolios to have a reasonable number of stocks.  We also ""hold"" for 6 months rather than 12, since most of the returns come in the first 6 months.","Keep only distcd 2nd digit = 2 or 3.  Define dividend initiation as having paid a dividend in month t (divamt > 0), and not having paid a dividend in the last 24 months.  DivInit is equal to 1 if a dividend was initiated in the past 6 months, and 0 for all other stocks.",1752.0
DivOmit,Predictor,1_clear,2_fair,"Michaely, Thaler and Womack",1995,Dividend Omission,JF,discrete,Event,payout indicator,1964,1988,DivOmit,t=6 in event study,"3 omit, to Day 254",event study 12 months,-1.0,0.916666667,6.33,EW,,,1.0,12.0,shrcd <= 11,"We deviate from OP in not imposing an NYSE/AMEX requirement.  This allows our portfolios to have a reasonable number of stocks.  OP's returns are short DivOmit and long EW crsp, which probably pushes up their t-stat.  Unlike DivInit, we ""hold"" for only 2 month because Table 3 shows that the DivOmit performance is highly concentrated early in the event.  Very few stocks in the omission portfolio.","Keep only distcd 2nd digit = 2 or 3.  Define firms as quarterly, semi-annual, or annual payers based on payment history.  Define a consistent payer as a firm that paid quarterly or semi-annual dividends for the past 18 months, or annual dividends for the past 2 years.  An omission is the first month where a consistent payer failed to pay a dividend over the past quarter/6-months/year.  Finally, DivOmit = 1 if there is an omission in the past 2 months and 0 otherwise",1752.0
DivSeason,Predictor,1_clear,1_good,Hartzmark and Salomon,2013,Dividend seasonality,JFE,discrete,Event,payout indicator,1927,2011,DivSeason,t=16 in long-short,2B long (1) short (2),LS port,1.0,0.36,16.19363096,EW,,,1.0,12.0,abs(prc)>5,,"Drop if 3rd digit of distcd = 2 or >= 6.  Assign DivSeason = 0 if there was a dividend paid in the last 12 months.  Replace DivSeason = 1 if the third digit of disctcd is 3, 0, or 1, and a positive dividend was paid 2, 5, 8, or 11 months ago, if the third digit is 4 and a dividend was paid 5 or 11 months ago, or if the third digit is 5 and a dividiend was paid 11 months ago.",154.0
DivYieldST,Predictor,1_clear,2_fair,Litzenberger and Ramaswamy,1979,Predicted div yield next month,JFE,discrete,Accounting,valuation,1936,1977,DivYieldST,t=6 in mv reg,1,mv reg,1.0,,6.3,EW,,,,,,"HXZ cite Litzenberger and Ramaswamy (LR) for their Dp predictor, but Dp is an annual dividend yield that is closer to Keim (1985), which shows very weak predictability.  LR is actually similar to DivSeason (Hartzmark and Solomon), and we follow LR.  LR uses a badly behaved regression with 75% of their div yield variable = 0, so we are flexible in our approach to mimic their results.  Also the paper is old, mostly theory, and provides little detail on their data handling.  Clear is a judgment call.","Using CRSP distributions, keep only distcd beginning in 12 and if the third digit of distcd is 3, 4, or 5.  Also keep only if stock paid a dividend in the last 12 months.  Define Ediv1 = div 2 months ago if distcd's 3rd digit is 0, 1, or 3.  Define Ediv1 = div 5 months ago if the distcd 3rd digit is 4.  Define Ediv = div 11 months ago if the distcd 3rd digit is 5.  Define Edy1 as Ediv1/abs(prc).  Finally, discretize Edy1 to smooth around the huge mass at 0 as follows: DivYieldST = 0 if Edy1 = 0, 1 if Edy is between 0 and 0.005, 2 if Edy1 is between 0.005 and 0.010, and 3 if Edy > 0.010.",2252.0
dNoa,Predictor,1_clear,1_good,"Hirshleifer, Hou, Teoh, Zhang",2004,change in net operating assets,JAE,continuous,Accounting,investment,1964,2002,dNoa,t=8.9 in mv reg,7B DeltaNOA,mv reg,-1.0,,8.85,EW,,,1.0,6.0,,Seems to be monthly.  Table 4 footnote says decile portfolios formed monthly for NOA.  Interestingly they have a minimum 4 month lag.,"12-month growth in Net Operating Assets scaled by lagged total assets (at). Net Operating assets are operating assets minus operating liabilities. Operating assets are total assets (at) minus cash- and short-term investments (che), operating liabilities are total assets minus long-term debt (dltt), minority interest (mib), deferred charges (dlc), book equity (ceq) and preferred stock (pstk), all items (except at and ceq) replaced with 0 if missing.",1136.0
DolVol,Predictor,1_clear,1_good,"Brennan, Chordia, Subra",1998,Past trading volume,JFE,continuous,Trading,volume,1966,1995,VolumeDol,t=2.9 in regression,6A,mv reg,-1.0,,2.86,EW,,,1.0,12.0,,"OP is an exploration style paper, and DolVol is just one predictor.  Tables 4-6 have related results, but Table 6 is the simplest.  Table 6A has t=2.86 for NYSE subsample, Table 6B has t=2.6 for NASDAQ subsample, we write down 2.86, but use NYSE and NASDAQ in our ports.",Log of two-month lagged trading volume (vol) times two-month lagged price (prc).,2219.0
EarningsConsistency,Predictor,1_clear,2_fair,Alwathainani,2009,Earnings consistency,BAR,continuous,Accounting,earnings growth,1971,2002,EarnCons,t=2.7 in complicated LS port,11A CLG-CHG,LS port,1.0,0.360833333,2.67,EW,,,12.0,6.0,abs(prc)>1,"Could not access OP, so we used the Alwathainani's dissertation from VCU.  We follow MP, which is simpler than OP.","Average earnings growth over previous 48 months. Earnings growth is defined as EPS (epspx) minus EPS 12 months ago divided by average EPS 12 and 24 months ago. Exclude if price less than 5, absolute value of 12 month earnings growth greater 600%, or earnings growth and earnings growth 12 months ago have different signs.",57.0
EarningsForecastDisparity,Predictor,1_clear,2_fair,Da and Warachka,2011,Long-vs-short EPS forecasts,JFE,continuous,Analyst,earnings forecast,1983,2006,LT_ST_EPS,t=5.1 in LS port,2B Month One,LS port,-1.0,0.48,5.08,EW,0.2,,1.0,12.0,,"Our strategy is simpler and follows HXZ. OP uses 3x3 sort, then LS corners.  OP t-stat is 4-factor alpha, but the factor premiums roughly cancel.",Analyst forecasted 5-year earnings growth (fgr5yr) minus 100 times the difference between mean earnings forecast (meanest) and fiscal year earnings expectations (fy0a) scaled by the absolute value of fy0a.  Drop if fpedats is missing or fpedats - statpers < 30,106.0
EarningsStreak,Predictor,1_clear,1_good,Loh and Warachka,2012,Earnings surprise streak,MS,continuous,Accounting,earnings growth,1987,2009,EarnStreak,t=9.5 in port sort ff3 alpha,3B Spread FF3 Streaks,port sort FF3 alpha,1.0,0.957,9.51,EW,0.2,,1.0,12.0,abs(prc)>5,"Table 3's footnote says stocks remain in the relevant portfolio for 6 months, which we take to mean that earnings announcements more than 6 months old are not used.  Announcements are quarterly anyway, so the rankings need to change quarterly.  We reassign portfolios monthly because earnings announcements occur throughout the year.  This was updated in 2021 February.  Old version was much simpler.",Use fpi == 6 and only the last statpers for each anndats_act.  Define surp = (actual - meanest)/price.  Define a firm-anndats as a streak if surp has the same sign as the most recent surp observation.  Keep only streaks.  Then define signal = surp.,92.0
EarningsSurprise,Predictor,1_clear,2_fair,"Foster, Olsen and Shevlin",1984,Earnings Surprise,AR,continuous,Accounting,earnings growth,1974,1981,EarnSurp,huge spread in event study,4 Days +1 to +60,event study 2 months,1.0,2.975,,EW,0.1,,1.0,12.0,abs(prc)>5,No LS but very strong return pattern,"EPS (epspxq) minus EPS twelve months ago - Drift, scaled by standard deviation of that expression. Drift is the average earnings growth (EPS - EPS twelve months ago) over the past two years. Exclude if price less than 5",1662.0
EarnSupBig,Predictor,1_clear,1_good,Hou,2007,Earnings surprise of big firms,RFS,continuous,Accounting,lead lag,1972,2001,EarnSupBig,t=9 in mv reg weekly,"AR_i,3",mv reg weekly,1.0,,8.91,EW,,,1.0,12.0,,Only shows up in Table 6.  t-stat is from firm-week regressions.,Average monthly value of EarningsSurprise (defined above) of the 30% largest companies by market value of equity in the same Fama-French 48 industry. Exclude the largest 30% of companies for EarnSupBig (not to compute the anomaly),814.0
EBM,Predictor,1_clear,1_good,"Penman, Richardson and Tuna",2007,Enterprise component of BM,JAR,continuous,Accounting,valuation,1963,2001,BMent,t=3.0 in double sort,4A NOA/P^NOA,double sort size adj,1.0,0.12,3.0,EW,0.1,,12.0,6.0,abs(prc)>5,"Enterprise = operating = NOA in OP.  So this should be NOA/P^NOA.  Need to adjust sample dates: even though paper says it begins in 1962, there is only 1 stock for the first 5 months of 1962. Table 4a has double sort with most t-stats above 3.  hand t-stat is 3.0 for simplicity and hand ret is 0.12.  OP drops extreme obs but we don't.",(ceq + che - dltt - dlc - dc - dvpa+ tstkp ) / (mve\_c + che - dltt - dlc - dc - dvpa+ tstkp). Exclude if price less than 5.,550.0
EntMult,Predictor,1_clear,1_good,Loughran and Wellman,2011,Enterprise Multiple,JFQA,continuous,Accounting,valuation,1963,2009,EntMult,t=6.54 in decile sort CAPM alpha,3B,port sort CAPM alpha,-1.0,0.95,6.54,EW,0.1,,12.0,6.0,,Table 3 Panel B. Table 3A shows raw returns but no t-stats.,"Market value of equity + long-term debt (dltt) + debt in current liabilities (dlc) + deferred charges (dc) - cash and short-term investments (che) , divided by operating income (oibdp). Exclude if missing book equity or negative operating income.",224.0
EP,Predictor,1_clear,1_good,Basu,1977,Earnings-to-Price Ratio,JF,continuous,Accounting,valuation,1957,1971,EP,monotonic port sort but no LS,1 average annual rate,"port sort, no LS",1.0,0.58,,EW,0.2,,1.0,6.0,exchcd==1,"MP use sample dates 1964-1971, but the original paper uses 1956-1971.  Also, original uses Dec 31 mve_c, so we lag by 6 months our monthly mve_c to approximate.","ib / lag(market value of equity, 6 months).  NYSE stocks only.  Exclude if EP < 0.  Lag simulates the Dec 31 market equity used in original paper",5357.0
EquityDuration,Predictor,1_clear,1_good,"Dechow, Sloan and Soliman",2004,Equity Duration,RAS,continuous,Accounting,valuation,1962,1998,Duration,t=4.4 in conservative long-short,6A HDMLD,LS FF3 style,-1.0,0.5,4.368488219,VW,0.2,,12.0,6.0,,"Tab6A uses FF93 style factor (HDMLD).  They don't seem to like the factor thing much and complain about it on page 14.  Our is just VW quintiles for simplicity,",see code,494.0
ExchSwitch,Predictor,1_clear,2_fair,Dharan and Ikenberry,1995,Exchange Switch,JF,discrete,Event,other,1962,1990,ExchSwitch,t = 3.6 in event study,2 Months-1-6,event study 6-month size adj,-1.0,0.455,3.61,EW,,,1.0,12.0,,"Fig 1 shows month 1 has most negative expected return.  Number of events approx 3,000, so this should work in portfolios.","Binary variable equal to 1 if a firm switched from AMEX or NASDAQ to NYSE within the past year, or from NASDAQ to AMEX within the past year.",385.0
ExclExp,Predictor,1_clear,1_good,"Doyle, Lundholm and Soliman",2003,Excluded Expenses,RAS,continuous,Analyst,composite accounting,1988,1999,ExcludExp,t=5.7 in mv reg,5 total exclusions RMA_1YR Full,mv reg,-1.0,,6.78,EW,,,12.0,6.0,,"Called total excusions in paper.  LS return is not at all monotonic, but reg (tab 5) is extremely significant.  Tab 7 port sort has a nonstandard p-value and subsets data, so we hand collect tab 5.",Difference between unadjusted earnings (EPSActualUnadj) from IBES and quarterly earnings per share (epspiq). Exclude the highest and lowest 1% of values.,741.0
FEPS,Predictor,1_clear,1_good,"Cen, Wei, and Zhang",2006,Analyst earnings per share,WP,continuous,Analyst,profitability,1983,2002,FEPS,t=2.7 in port sort,2 Ret0:1,port sort,1.0,1.199,2.66,EW,0.1,,1.0,12.0,abs(prc)>5,,"Using IBES unadjusted forecasts, keep fpi == 1, signal is meanest.",13.0
fgr5yrLag,Predictor,1_clear,1_good,La Porta,1996,Long-term EPS forecast,JF,continuous,Analyst,earnings forecast,1983,1990,EPSForeLTlag,t=4.9 in regression,3 E{g},univariate reg,-1.0,,4.9,EW,,,3.0,6.0,,"Port sort in Tab 2 is very strong but no t-stats.  Regression in Tab 3 E{g} is univariate.  We find the timing of the lag is important, but it may be more important to compound monthly returns to annual, as described on page 1717.  If we compound returns, the high volatility of the high long-term-growth stocks leads to very poor performance, consistent with Table II of OP and Figure 1 of the 2019 JF paper.  We are grateful to Rafael La Porta for helping us identify this feature.  But for our paper, our portfolio is just the arithmetic mean of monthly returns for simplicity.","Lag long-term earnings forecast (fgr5yr) by 6 months.   Then keep only June observations, and fill in missing with most recent obs.  Exclude if book equity (ceq), net income (ib), deferred taxes (txdi), dividends (dvp), revenue (sale) or depreciation (dp) is missing.  Keep only",1468.0
FirmAgeMom,Predictor,1_clear,1_good,Zhang,2006,Firm Age - Momentum,JF,continuous,Price,momentum,1983,2001,MomYoung,t = 7.21 in long portfolio,4 middle U5,port sort,1.0,2.9,7.21,EW,0.2,,1.0,6.0,,"Returns are monotonic in momentum, suggesting this variable can be continuous.","6 month return, restricted to the bottom quintile of the cross-sectional firm age distribution. Exclude if price less than 5 or firm younger than 12 months.",2599.0
ForecastDispersion,Predictor,1_clear,1_good,"Diether, Malloy and Scherbina",2002,EPS Forecast Dispersion,JF,continuous,Analyst,volatility,1976,2000,EPSDisp,t=2.9 in port sort,2,port sort,-1.0,0.79,2.88,EW,0.2,,1.0,6.0,,,Keep fpi = 1 and fpedats > statpers + 30.  Standard deviation of earnings estimates (stdev\_est) scaled by mean earnings estimate.,2775.0
Frontier,Predictor,1_clear,1_good,Nguyen and Swanson,2009,Efficient frontier index,JFQA,continuous,Accounting,valuation,1980,2003,EffFrontier,t=5 in port sort,4A Spread,port sort,1.0,0.96,4.863206041,EW,0.1,,12.0,6.0,"exchcd%in%c(1,2,3),shrcd<=11",,"Frontier is the residual of a regression of log(BM) on log(book equity (ceq)), long-term debt (dltt) to assets (at), capital expenditures (capx) to revenue (sale), R&D expense (xrd) to revenue, advertising expense (xad) to revenue, property plant and equipment (ppent) to assets, EBIT (ebitda) to assets, and dummies for Fama-French's 48 industry definitions. Regression is updated each month with a rolling window of 60 months.",148.0
Governance,Predictor,1_clear,1_good,"Gompers, Ishii and Metrick",2003,Governance Index,QJE,discrete,Other,other,1990,1999,Governance,t=2.7 in long short FF3 alpha,7 (1),,-1.0,0.72,2.769230769,VW,,,1.0,12.0,,"Tab7 shows value-weighted FF alpha t-stat of 2.73. No loadings provided.  Table 6 shows portfolio FF3 regs, and that both SMB and HML loadings contribute to spread, so it's not super clear if the raw LS will be significant. Port sort is not very monotonic.","Index available from http://faculty.som.yale.edu/andrewmetrick/data.html . The index is only available every 2-3 years for each firm, we replace intermediate missing values with the latest available one.  Value-weighted.",12361.0
GP,Predictor,1_clear,1_good,Novy-Marx,2013,gross profits / total assets,JFE,continuous,Accounting,profitability,1963,2010,ProfGross,t=2.5 in VW LS quint,2a,port sort,1.0,0.31,2.49,VW,0.2,,12.0,6.0,,"Tab 2a says NYSE breakpoints, but our code gets much closer to their result without all stock breakpoints.","Revenue (sale) - cost of goods solds (cogs), divided by  total assets (at).  Drop if financial.",2980.0
GrAdExp,Predictor,1_clear,1_good,Lou,2014,Growth in advertising expenses,RFS,continuous,Accounting,investment alt,1974,2010,AdExpGr,t=3.5 in long-short,2A Year 1 Excess,LS port,-1.0,0.58,3.54,EW,0.1,,1.0,12.0,abs(prc)>5,"Table 2, panel A , 10-1.  Portfolios are ""rebalanced every month using the most recent advertising spending data""","Log of advertising expense (xad) minus log of advertising expense last year. Exclude if price less than 5, xad less than .1 or stock in the lowest decile of market value of equity.",629.0
grcapx,Predictor,1_clear,1_good,Anderson and Garcia-Feijoo,2006,Change in capex (two years),JF,continuous,Accounting,investment growth,1976,1999,CAPXgr,t=5 in port sort,3B cegth2,port sort,-1.0,0.57,5.05,EW,0.2,,12.0,6.0,,called cegth2,"Growth rate of capital expenditures (capx) relative to two years ago.   If capx is missing, replace with annual change in property, plant and equipment (ppent).",587.0
grcapx3y,Predictor,1_clear,1_good,Anderson and Garcia-Feijoo,2006,Change in capex (three years),JF,continuous,Accounting,investment growth,1976,1999,CAPXgr3y,t=4.7 in port sort,3D cegth3,port sort,-1.0,0.6,4.71,EW,0.2,,12.0,6.0,,"We follow OP, not HXZ.  OP notation is odd, uses cegth2 and cegth3 but no cegth.","Capital expenditures (capx) divided by the sum of capital expenditures from year - 1, year -2, and year -3.   If capx is missing, replace with annual change in property, plant and equipment (ppent).",587.0
Herf,Predictor,1_clear,1_good,Hou and Robinson,2006,Industry concentration (sales),JF,continuous,Other,other,1963,2001,Herf,t = 2.14 in port sort,2 firm-level raw,port sort,-1.0,0.26,2.14,EW,0.2,,1.0,6.0,,"Tab3 Raw t-stat of 2.14. Table 3, firm-level return set of columns","Three-year rolling average of the three digit industry Herfindahl index based on firm revenue (sale). Exclude regulated industries (4011, 4210, 4213 & year $\leq$ 1980; 4512 & year $\leq$ 1978, 4812, 4813 & year $\leq$ 1982, 4900-4999 in any year)",1316.0
HerfBE,Predictor,1_clear,1_good,Hou and Robinson,2006,Industry concentration (equity),JF,continuous,Other,other,1963,2001,HerfBE,t = 2.52 in characteristics-adjusted port sort,2 H(Equity),port sort char adjusted,-1.0,0.24,2.52,EW,0.2,,1.0,6.0,,Tab3 H(Equity) characteristic adjusted t-stat 2.52. Judgment call.,"Three-year rolling average of the three digit industry Herfindahl index based on firm book equity. Exclude regulated industries (4011, 4210, 4213 & year $\leq$ 1980; 4512 & year $\leq$ 1978, 4812, 4813 & year $\leq$ 1982, 4900-4999 in any year)",1316.0
hire,Predictor,1_clear,1_good,"Bazdresch, Belo and Lin",2014,Employment growth,JPE,continuous,Other,investment alt,1965,2010,LaborGr,t=5.8 in port sort,1A,port sort,-1.0,0.87,5.78,EW,0.2,,12.0,6.0,,,"Change in number of employees (emp) between t -1 and t, scaled by average number of employees in t-1 and t. Replace hire with 0 if emp or lagged emp is missing.",489.0
IdioVol3F,Predictor,1_clear,1_good,Ang et al.,2006,Idiosyncratic risk (3 factor),JF,continuous,Price,volatility,1963,2000,IdioVol3F,t=3.1 in port sort,7B,port sort FF3 alpha,-1.0,1.06,3.1,VW,0.2,,1.0,6.0,,,Standard deviation of residuals from Fama-French three factor regressions using the past month of daily data.  Value weighted,6323.0
IdioVolAHT,Predictor,1_clear,1_good,"Ali, Hwang, and Trombley",2003,Idiosyncratic risk (AHT),JFE,continuous,Price,volatility,1976,1997,IdioVolAHT,t = 2.7 in mv reg,4 Ivolatility^-1,reg,-1.0,,2.699,VW,,,36.0,6.0,,"dep var is size adjusted return, to mimic, we value weight our ports.  This result is really buried in the paper.  Most all of the paper is on how B/M predictatbility varies by subsets of stocks.",Standard deviation of residuals from CAPM regressions using the past year of daily data. Require at least 100 non-missing observations.,748.0
Illiquidity,Predictor,1_clear,1_good,Amihud,2002,Amihud's illiquidity,JFM,continuous,Trading,liquidity,1964,1997,Illiquid,t=6.6 in univariate reg,2,reg,1.0,,6.6,EW,,,12.0,6.0,"abs(prc)>5, exchcd==1",,Past twelve month average of: daily return (abs(ret)) divided by turnover((abs(prc)*vol),14669.0
IndIPO,Predictor,1_clear,2_fair,Ritter,1991,Initial Public Offerings,JF,discrete,Event,external financing,1975,1987,IndIPO,t=4 in event study,2 month 12,event study 12 month firm match,-1.0,0.8525,3.97,EW,,,1.0,12.0,,"Tab II event study t-stat 5 at 1 year.  Slightly different than MP.  Our construction is closest in the spirit of Ritter's Table II.  A few issues.  Main thing: sample should end in 1987, since the main table is for returns of stocks which IPO'd 1975-1984, and the table follows the stocks for 3 years (past 1984).  Also: using minimum of 3 months since IPO helps.",1 if IPO in the past 6-36 months.  0 otherwise.  IPO dates are taken from Jay Ritter's IPO data available at: http://bear.warrington.ufl.edu/ritter/ipodata.htm.  Missing IPO dates imply IndIPO = 0,936.0
IndMom,Predictor,1_clear,1_good,Grinblatt and Moskowitz,1999,Industry Momentum,JF,continuous,Price,momentum,1963,1995,IndMom,t=4.6 in long-short,2A Raw,LS port nonstandard (industry),1.0,0.43,4.65,EW,0.3,,6.0,6.0,,"More results on Table 3.  OP's long port equally weights three industry portfolios, where the three industries are the top 3 according to the signal and the industry portfolios are value-weighted (similarly for the short port).  We approximate this by equally weighting stocks.  Performance is weaker, reminiscent of our Menzly and Ozbas replication.",Weighted average of firm-level 6 month buy-and-hold return. Average is taken over two digit industries each month and weights are based on market value of equity.,2565.0
IndRetBig,Predictor,1_clear,1_good,Hou,2007,Industry return of big firms,RFS,continuous,Price,lead lag,1972,2001,IndRetBig,t=11 in mv reg,"6 R_i,3(-1)",mv reg weekly,1.0,,11.0,EW,,,1.0,12.0,,Table 2 presents a VAR with two states: return of big firms and return of small firms.  Table 6 is easier to compare to others.,Average monthly return (ret) of the 30% largest companies by market value of equity in the same Fama-French 48 industry. Exclude the largest 30% of companies for IndRetBig (not to compute the anomaly!),814.0
IntanBM,Predictor,1_clear,1_good,Daniel and Titman,2006,Intangible return using BM,JF,continuous,Accounting,long term reversal,1968,2003,IntanBM,t=4.0 in mv reg,4 reg 3 r^I(B),mv reg,-1.0,,3.99,EW,,,1.0,6.0,abs(prc)>5,,"In each month, run a cross-sectional regression of a firm's five-year stock return on 5 year lagged log BM (book equity (ceq)/market value of equity) and a constructed regressor that is the change in log BM from 5 years ago to today plus the five-year stock return. The residual from that regression is IntanBM.",1640.0
IntanCFP,Predictor,1_clear,1_good,Daniel and Titman,2006,Intangible return using CFtoP,JF,continuous,Accounting,long term reversal,1968,2003,IntanCFP,t=4.9 in mv reg,4 r^I(C),mv reg,-1.0,,4.89,EW,,,1.0,6.0,abs(prc)>5,,"In each month, run a cross-sectional regression of a firm's five-year stock return on the 5 year lagged log CFP, where CFP = (income before extraordinary items (ib) plus depreciation (dp))/market value of equity, and a constructed regressor that is the change in log CFP from 5 years ago to today plus the five-year stock return. The residual from that regression is IntanCFP.",1640.0
IntanEP,Predictor,1_clear,1_good,Daniel and Titman,2006,Intangible return using EP,JF,continuous,Accounting,long term reversal,1968,2003,IntanEP,t=4.6 in mv reg,4 r^I(E),mv reg,-1.0,,4.64,EW,,,12.0,6.0,abs(prc)>5,,"In each month, run a cross-sectional regression of a firm's five-year stock return on the 5 year lagged log EP, where EP = income before extraordinary items (ib)/market value of equity, and a constructed regressor that is the change in log EP from 5 years ago to today plus the five-year stock return. The residual from that regression is IntanEP.",1640.0
IntanSP,Predictor,1_clear,2_fair,Daniel and Titman,2006,Intangible return using Sale2P,JF,continuous,Accounting,long term reversal,1968,2003,IntanSP,t=4.3 in mv reg,4 r^I(S),mv reg,-1.0,,4.24,EW,,,12.0,6.0,,Not as clear as the other Intan* variables because the other coefficients in the regression are significant.  We took the liberty to remove the price filter and categorize this as clear,"In each month, run a cross-sectional regression of a firm's five-year stock return on 5 year lagged log SP (sales (sale)/market value of equity) and a constructed regressor that is the change in log SP from 5 years ago to today plus the five-year stock return. The residual from that regression is IntanSP.",1642.0
IntMom,Predictor,1_clear,1_good,Novy-Marx,2012,Intermediate Momentum,JFE,continuous,Price,momentum,1927,2010,Mom12to7,Tab2 t-stat 5.79,2 column 1,port sort,1.0,1.2,5.79,VW,0.1,,1.0,6.0,,"Text says they use VW, but EW performs similarly. Table 2, column (1).",Stock return between months t-12 and t-6,696.0
Investment,Predictor,1_clear,1_good,"Titman, Wei and Xie",2004,Investment to revenue,JFQA,continuous,Accounting,investment,1973,1996,InvToRev,t=2.86 in VW port sort,1B Average,"port sort, characteristic adjusted",-1.0,0.168333333,2.86,VW,0.2,,12.0,6.0,,OP mean return is only 17 bps per month characteristic adjusted.  We deviate somewhat from OP in the port sort by going LS 5-1 instead of (4+5) - (1+2) because OP value weights within each quintile which we can't do easily in our code.,Ratio of capital investment (capx) to revenue (revt) divided by the firm-specific 36-month rolling mean of that ratio. Exclude if revenue less than \$10m.,2415.0
InvestPPEInv,Predictor,1_clear,1_good,"Lyandres, Sun and Zhang",2008,change in ppe and inv/assets,RFS,continuous,Accounting,investment,1970,2005,InvestPPEInv,t=7 in long-short port,text p 2837,LS port nonstandard,-1.0,0.57,7.13,EW,0.3,,1.0,6.0,,"Not in a table.  Page 2837 has untabulated results.  OP does a complicated 3x3 sort with size two stage portfolio construction, but we just equal-weight.","One-year change in property, plants and equipment (ppegt) plus one year change in inventory (invt), scaled by one-year lagged assets (at).",665.0
InvGrowth,Predictor,1_clear,1_good,Belo and Lin,2012,Inventory Growth,RFS,continuous,Accounting,profitability,1965,2009,InvenGr,t=6.6 in port sort,2A EW,port sort,-1.0,0.89,6.64,EW,0.1,,12.0,6.0,,,"Defate invt growth using gnp deflator.  Signal is deflated invt growth rate from fiscal year t to fiscal year t-1.  Drop if 1 digit sic code is 4 or 6, or if at or ppent <= 0",220.0
iomom_cust,Predictor,2_likely,2_fair,Menzly and Ozbas,2010,Customers momentum,JF,continuous,Other,lead lag,1986,2005,iomom_cust,t=4 in mv reg,"2 (1) r_customer,t-1",mv reg,1.0,,4.11,EW,,,1.0,6.0,,"Early in the paper, they use a stock level  mv reg with many controls.  Later in the paper, they sort industry portfolios instead of stocks.  Given that it would have been more straightforward to just sort on stocks, I think the evidence on standard portfolio sorts is unclear (hence this is a likely predictor, not clear)","We download BEA Make-Use Tables, match to Compustat by NAICS assuming 5 year lag, compute returns within BEA 70 industries.  Then match each industry-year to matched-industry weights corresponding to IO tables as follows: for supplier mom, industry comes from cols of Use table, matched industries from the rows.  For customer mom: industry from rows of make table, matched industries from cols.  Weights exclude own-industry entries.  For each industry-month, average returns of matched industries, weighted using IO tables.  Sort industries into deciles based on matched returns, assign firm-months to industry deciles, Finally, long stocks in deciles 8-10 and short decile 1. Drop pre-1986 due to NAICS availability.  ",727.0
iomom_supp,Predictor,2_likely,2_fair,Menzly and Ozbas,2010,Suppliers momentum,JF,continuous,Other,lead lag,1986,2005,iomom_supp,t=5 in mv reg,"2 (1) r_supplier,t-1",mv reg,1.0,,5.03,EW,,,1.0,6.0,,"Early in the paper, they use a stock level  mv reg with many controls.  Later in the paper, they sort industry portfolios instead of stocks.  Given that it would have been more straightforward to just sort on stocks, I think the evidence on standard portfolio sorts is unclear (hence this is a likely predictor, not clear)","We download BEA Make-Use Tables, match to Compustat by NAICS assuming 5 year lag, compute returns within BEA 70 industries.  Then match each industry-year to matched-industry weights corresponding to IO tables as follows: for supplier mom, industry comes from cols of Use table, matched industries from the rows.  For customer mom: industry from rows of make table, matched industries from cols.  Weights exclude own-industry entries.  For each industry-month, average returns of matched industries, weighted using IO tables.  Sort industries into deciles based on matched returns, assign firm-months to industry deciles, Finally, long stocks in deciles 8-10 and short decile 1. Drop pre-1986 due to NAICS availability.  ",727.0
Leverage,Predictor,1_clear,1_good,Bhandari,1988,Market leverage,JF,continuous,Accounting,leverage,1952,1981,Leverage,t=3.9 in regression,1 DER,mv reg,1.0,,3.93,EW,,,12.0,6.0,,,Total liabilities (lt) divided by market value of equity.,2597.0
LRreversal,Predictor,1_clear,1_good,De Bondt and Thaler,1985,Long-run reversal,JF,continuous,Price,long term reversal,1929,1982,Mom36m,t=3.3 in long-short,1 three-year  1 month,LS port CAPM alpha,-1.0,0.105,3.29,EW,,,1.0,12.0,,Insignificant at 12-month horizon.  Many alternative signal designs lead to similar results.,Stock return between months t-36 and t-13.,13447.0
MaxRet,Predictor,1_clear,1_good,"Bali, Cakici, and Whitelaw",2011,Maximum return over month,JFE,continuous,Price,volatility,1962,2005,MaxRet,t=2.8 in port sort,1 VW 10-1,port sort,-1.0,1.03,2.83,VW,0.1,,1.0,6.0,,Doesn't work as well EW.  Very nonlinear return vs decile,Maximum of daily returns (ret) over the previous month,2069.0
MeanRankRevGrowth,Predictor,1_clear,2_fair,"Lakonishok, Shleifer, Vishny",1994,Revenue Growth Rank,JF,continuous,Accounting,sales growth,1968,1990,RevGrowth,t=4.5 in double sort,6 panel 2,LS port,1.0,,,EW,0.2,,12.0,6.0,"exchcd%in%c(1,2)","Lots of supporting results, but not exactly what we do.  Tab 6 panel 2 finds t=4.5 using 3x3 sort with CF and LS corners.","Rank firms by their annual revenue growth each year over the past 5 years. MeanRankRevGrowth is the weighted average of ranks over the past 5 years, that is, MeanRankRevGrowth = (5*Rank$_{t-1}$ + 4*Rank$_{t-2}$ + 3*Rank$_{t-3}$ + 2*Rank$_{t-4}$ + 1*Rank$_{t-5}$)/15. Exclude NASDAQ stocks.",7664.0
Mom12m,Predictor,1_clear,1_good,Jegadeesh and Titman,1993,Momentum (12 month),JF,continuous,Price,momentum,1964,1989,Mom12m,t=3.7 long-short,1A K=3 row 12,port sort,1.0,1.31,3.74,EW,0.1,,3.0,6.0,,see Mom6m,Stock return between months t-12 and t-1,17341.0
Mom12mOffSeason,Predictor,1_clear,1_good,Heston and Sadka,2008,Momentum without the seasonal part,JFE,continuous,Price,other,1965,2002,Mom12mOffSeason,t=4 in port sort,2 Year 1 Nonannual,port sort,1.0,1.17,4.2,EW,0.1,,1.0,6.0,"exchcd%in%c(1,2)",This acronym has a different form than the other off season Heston and Sadka ones because its behavior is distinct.  The other off season signals behave like long-term reversal.,Average return in  other months over the previous year.,442.0
Mom6m,Predictor,1_clear,1_good,Jegadeesh and Titman,1993,Momentum (6 month),JF,continuous,Price,momentum,1964,1989,Mom6m,t=2.4 long-short,1A K=3 row 6,port sort,1.0,0.84,2.44,EW,0.1,,3.0,6.0,,"OP doesn't skip the current month in the signal, and instead skips one week before buying the portfolio.  We follow McLean and Pontiff, Green-Hand-Zhang, and others in skipping the current month instead.",Stock return between months t-6 and t-1,17341.0
Mom6mJunk,Predictor,1_clear,1_good,Avramov et al,2007,Junk Stock Momentum,JF,continuous,Price,momentum,1985,2003,Mom6Jnk,t=4.3 in port sort,3 NIG,port sort,1.0,2.12,4.29,EW,0.1,,1.0,6.0,"abs(prc)>5, me>me_nyse20",,Mom6m. Include only stocks with a credit rating (splticrm) of BBB or lower. Drop if missing credit rating or non-standard credit rating.,593.0
MomOffSeason,Predictor,1_clear,1_good,Heston and Sadka,2008,Off season long-term reversal,JFE,continuous,Price,other,1965,2002,MomOffSeason,t=5.6 in port sort,2 Years 2-5 Nonannual,port sort,-1.0,1.25,5.6,EW,0.1,,1.0,6.0,"exchcd%in%c(1,2)",,Average return in other months over the preceding 2-5 years.,442.0
MomOffSeason06YrPlus,Predictor,1_clear,1_good,Heston and Sadka,2008,Off season reversal years 6 to 10,JFE,continuous,Price,other,1965,2002,MomOffSeason6YrPlus,t=4.6 in port sort,2 Years 6-10 Nonannual,port sort,-1.0,0.55,4.62,EW,0.1,,1.0,6.0,"exchcd%in%c(1,2)",,Average return in other months over the preceding 6-10 years.,442.0
MomOffSeason16YrPlus,Predictor,1_clear,1_good,Heston and Sadka,2008,Off season reversal years 16 to 20,JFE,continuous,Price,other,1965,2002,MomOffSeason16YrPlus,t=3.4 in port sort,2 Years 16-20 nonannual,port sort,-1.0,0.39,3.35,EW,0.1,,1.0,6.0,"exchcd%in%c(1,2)",,Average return in other months over the preceding 16-20 years.,442.0
MomRev,Predictor,1_clear,1_good,Chan and Ko,2006,Momentum and LT Reversal,JOIM,discrete,Price,momentum,1965,2001,MomRev,t=4.3 in long-short,5,LS port FF3 alpha,1.0,0.48,4.29,EW,,,3.0,6.0,,,"Binary variable equal to 1 if firm is in the highest Mom6m quintile and the lowest Mom36m quintile, and equal to 0 if firm is in the lowest Mom6m quintile and the highest Mom36m quintile. Exclude if price less than 5.",17.0
MomSeason,Predictor,1_clear,1_good,Heston and Sadka,2008,Return seasonality years 2 to 5,JFE,continuous,Price,other,1965,2002,MomSeason,t=5 in port sort,2 Years 2-5 Annual,port sort,1.0,0.67,5.35,EW,0.1,,1.0,6.0,"exchcd%in%c(1,2)","Perhaps one should only include this signal from Heston and Sadka.  Certainly one shouldn't have 5 flavors of seasonal momentum, and on top of that 5 flavors of the off-season.  We include all of this to nest HXZ, however.  And technically, all of these strategies are found in the paper.",Average return in the same month over the preceding 2-5 years.,442.0
MomSeason06YrPlus,Predictor,1_clear,1_good,Heston and Sadka,2008,Return seasonality years 6 to 10,JFE,continuous,Price,other,1965,2002,MomSeason06YrPlus,t=6.1 in port sort,2 Years 6-10 Annual,port sort,1.0,0.68,6.15,EW,0.1,,1.0,6.0,"exchcd%in%c(1,2)",,Average return in the same month over the preceding 6-10 years.,442.0
MomSeason11YrPlus,Predictor,1_clear,1_good,Heston and Sadka,2008,Return seasonality years 11 to 15,JFE,continuous,Price,other,1965,2002,MomSeason11YrPlus,t=6.4 in port sort,2 Years 11-15 nonannual,port sort,1.0,0.66,6.43,EW,0.1,,1.0,6.0,"exchcd%in%c(1,2)",,Average return in the same month over the preceding 11-15 years.,442.0
MomSeason16YrPlus,Predictor,1_clear,1_good,Heston and Sadka,2008,Return seasonality years 16 to 20,JFE,continuous,Price,other,1965,2002,MomSeason16YrPlus,t=4.5 in port sort,2 Years 16-20 annual,port sort,1.0,0.52,4.58,EW,0.1,,1.0,6.0,"exchcd%in%c(1,2)",,Average return in the same month over the preceding 16-20 years.,442.0
MomSeasonShort,Predictor,1_clear,1_good,Heston and Sadka,2008,Return seasonality last year,JFE,continuous,Price,other,1965,2002,MomSeasonShort,t=7.6 in port sort,2 Year 1 Annual,port sort,1.0,1.15,7.6,EW,0.1,,1.0,6.0,"exchcd%in%c(1,2)",,Average return in the same monthin the previous year.,442.0
MomVol,Predictor,1_clear,1_good,Lee and Swaminathan,2000,Momentum in high volume stocks,JF,discrete,Price,momentum,1965,1995,MomVol,"t=6 in long-short, lots of robustness",2 J=6 K=3 V3 R10-R1,port sort,1.0,1.55,5.78,EW,0.2,,3.0,6.0,"abs(prc)>1, exchcd %in% c(1,2)",We use monthly instead of daily volume.,"Define momentum as Mom6m, and volume as the rolling average of the past 6 months of monthly turnover (minimum 5 months).  Independent sort stocks into 10 momentum ports and 3 volume ports. Keep if volume is in the top port, and assign signal = momentum port.  Drop if less than 2 years on CRSP.",2509.0
MS,Predictor,1_clear,2_fair,Mohanram,2005,Mohanram G-score,RAS,discrete,Accounting,composite accounting,1978,2001,Mscore,t=9 in port sort nonstandard data lag,4A,port sort,1.0,1.575,9.14,EW,,,1.0,12.0,,"OP's signal is really complicated, and the text is not extremely detailed.  We do the best we can to mimic OP's results, but complications in data lagging, combining annual and quarterly data, and sample selection make this extremely difficult.  We get a t-stat near 6 and monotonic returns, quite in the spirite of OP, but still far from OP's t-stat of 9.  Could be that OP uses overlapping samples.  We use Portfolio Period = 1 to try to get close.","See code for details.  MS is only evaluated for low BM firms and comes from combining three signals related to profitability and cash flow, two signals related to income volatility, and three signals related to investment.",561.0
NetDebtFinance,Predictor,1_clear,1_good,"Bradshaw, Richardson, Sloan",2006,Net debt financing,JAE,continuous,Accounting,external financing,1971,2000,NDebtFin,t=6.9 in port sort,3,port sort size adjusted,-1.0,0.675,6.91,EW,0.1,,12.0,6.0,,,"Long-term debt issuance (dltis) minus long-term debt reduction (dltr) minus current debt changes (dlcch), scaled by average total assets (at) in years t-1 and t. Replace missing values of dlcch with 0. Exclude if ratio is greater than 1.",798.0
NetDebtPrice,Predictor,1_clear,1_good,"Penman, Richardson and Tuna",2007,Net debt to price,JAR,continuous,Accounting,leverage,1963,2001,NDebtPrice,t=2.3 in double sort,4A,double sort size adjusted,-1.0,0.658333333,2.34,EW,0.2,,12.0,6.0,,"need to adjust sample dates: even though paper says it begins in 1962, there is only 1 stock for the first 5 months of 1962.  ND/P in paper.  Table 4a has size adjusted returns for double sorts, on NDP and NOA/POA.  I chose 5th NOA/POA Portfolio, but they're all similar.  Table 1 Panel C has decile sort returns, but no hedge returns.  Performance here is not great, as noted in the text.  Best performance is Table 4 which reports conditional sorts.  OP drops extreme obs but we don't.","Long-term debt (dltt) plus debt in current liabilities (dlc) plus preferred stock (pstk) plus preferred dividends in arrears (dvpa) minus treasury stock (tstkp) minus cash and short-term investments (che), scaled by market value of equity. Exclude if SIC between 6000 and 6999, or if missing value for total assets (at), net income (ib), common shares outstanding (csho), book value of equity (ceq) or price close fiscal year (prcc\_f).  Keep only 3rd B/M Quintile, following Table 4 (and in contrast to Table 1).",550.0
NetEquityFinance,Predictor,1_clear,1_good,"Bradshaw, Richardson, Sloan",2006,Net equity financing,JAE,continuous,Accounting,external financing,1971,2000,NEqFin,t=3.8 in port sort,3,port sort size adjusted,-1.0,0.93,3.82,EW,0.1,,12.0,6.0,,,"Sale of common stock (sstk) minus  purchase of common stock (prstkc), scaled by average total assets (at) from years t and t-1. Exclude if absolute value of ratio is greater than 1.",798.0
NetPayoutYield,Predictor,1_clear,2_fair,Boudoukh et al.,2007,Net Payout Yield,JF,continuous,Accounting,valuation,1984,2003,NPayYield,nearly monotonic NYSE decile sort but no LS t-stat,III D,port sort no LS,1.0,0.73,,EW,0.1,NYSE,12.0,6.0,,"Table III D NYSE deciles of nonzero yields; 0.73 is D10 minus D1, no t-stat in OP. Table VI B 30/70 sort (FF3 alpha t=2.83) is conservative and spec-sensitive; ours gives t=1.27 raw. Our deciles track Table III closely. See issue #207.","Gross payout is dividends (dvc) plus purchase of common and preferred stock (prstkc) plus max(pstkrv lagged 12 months minus pstkrv, 0). Equity issuance is sale of common and preferred stock (sstk) minus max(pstkrv minus pstkrv lagged 12 months, 0). NetPayoutYield is gross payout minus equity issuance, divided by mve_permco lagged 6 months. An unavailable preferred-stock difference contributes zero to both adjustments; missing flow inputs remain missing. Require finite nonzero yield and finite positive lagged market cap. Exclude financial firms based on SIC code, missing or nonpositive ceq, or less than 2 years in CRSP.",777.0
NOA,Predictor,1_clear,1_good,Hirshleifer et al.,2004,Net Operating Assets,JAE,continuous,Accounting,asset composition,1964,2002,NOA,t=8.5 in long-short,4 H-L,LS port size+BM+Mom adjusted,-1.0,1.48,8.45,EW,0.1,,1.0,6.0,,,"Difference between operating assets and operating liabilities, scaled by lagged total assets. Operating assets are total assets (at) minus cash- and short-term investments (che), operating liabilities are total assets minus long-term debt (dltt), minority interest (mib), deferred charges (dc) and book equity (ceq).",1138.0
OPLeverage,Predictor,1_clear,1_good,Novy-Marx,2011,Operating leverage,ROF,continuous,Accounting,other,1963,2008,OperLeverage,t=3.38 in port sort,3b,port sort,1.0,0.51,3.38,EW,0.2,,12.0,6.0,,"Table III panel b shows EW raw returns strong (t-stat 3.38), somewhat weaker after factor adjustment or in VW returns. Table IIIv has both raw and alphas.","Sum of administrative expenses (xsga) and cost of goods sold (cogs), scaled by total assets (at).  Use xsga = 0 if xsga is missing.",534.0
OptionVolume1,Predictor,1_clear,2_fair,Johnson and So,2012,Option to stock volume,JFE,continuous,Options,volume,1996,2010,OptVol,t = 3.45 in port sort CAPM alpha weekly,2A (O/S),port sort CAPM alpha nonstandard,-1.0,0.916,3.45,EW,0.2,,1.0,12.0,"abs(prc)>1, shrcd<=11","OP is actually weekly, and sorts in deciles, then  longs 9+10 and shorts 1+2","Total monthly option volume (volume) over all puts and calls, divided by monthly stock trading volume (vol). Exclude if price less than 1 or share code greater 11 or option volume or stock volume data are missing for the previous month.",460.0
OrderBacklog,Predictor,1_clear,1_good,"Rajgopal, Shevlin, Venkatachalam",2003,Order backlog,RAS,continuous,Accounting,sales growth,1981,1999,OrderBacklog,t=2.38 in univariate size-adjusted FMB,3A gamma1,univariate reg,-1.0,,2.38,EW,,,12.0,6.0,,"Table 3, but FMB size adjusted only. Other tables use nonlinear regressions.",Order backlog (ob) divided by average total assets (at) in years t-1 and t. Exclude if order backlog is 0.,295.0
OrderBacklogChg,Predictor,1_clear,2_fair,Baik and Ahn,2007,Change in order backlog,Other,continuous,Accounting,accruals,1971,1999,OrderBacklogChg,p<0.01 in port sort,2 High-Low,port sort,1.0,1.1675,,EW,0.1,,12.0,6.0,,OP is more aggressive than most and assumes a 4 month lag between fiscal year end and data availability.  We do the standard 6 month.,Define normalized order backlog  as order backlog (ob) divided by average total assets (at) in years t-1 and t. Exclude if order backlog is 0.  Signal is normalized order backlog minus normalized order backlog one year ago.,23.0
OrgCap,Predictor,1_clear,1_good,Eisfeldt and Papanikolaou,2013,Organizational capital,JF,continuous,Accounting,R&D,1970,2008,OrgCap,t=2.9 in port sort,4A.1,port sort less rf,1.0,0.385833333,2.85,VW,0.2,,12.0,6.0,,OP only examines the industry adjusted version and explains clearly why.,"Defined recursively. Set xsga to 0 if missing.  Initialize with OrgCap = 4*general expenses (xsga) in the first year, and calculate as .85*OrgCap previous year + xsga/gnpdeflator current year thereafter. Scale by total assets (at).  Keep only december year ends, and non-financials, and non-missing sic code.  Then winsoriz at the 1% level and standardized by subtracting the same month within-industry mean (Fama French 17 industries) and dividing by the same month within industry standard deviation of OrgCap.",1211.0
OScore,Predictor,1_clear,1_good,Dichev,1998,O Score,JF,discrete,Accounting,default risk,1981,1995,OScore,t=3.36 in LS port,5 highest 10 percent,LS port,-1.0,1.17,3.36,EW,,,1.0,6.0,abs(prc)>5,"Table 4 has returns by portfolio, but no t-stats.  Table 5 shows t=3.36 if you long low 70% and short high 10%.  OP does not mention price screen, but without the screen results are far, and with it results are very close.","OScore = -1.32 - .407*log(at/GNP deflator) + 6.03*(lt/at) - 1.43*( (act - lct)/at) + .076*(lct/act) - 1.72*I(lt > at) - 2.37*(ib/at) - 1.83*(fopt/lt) + .285*(ib + ib$_{t-12}$ + ib$_{t-24}$ < 0) -  .521*( (ib - ib$_{t-12}$)/(abs(ib) + .abs(ib$_{t-12}$)) ). fopt = oancf if fopt is missing.  Exclude Exclude if SIC code between 3999 and 4999, or greater than 5999. Exclude if price less than 5.  Then exclude if OScore is in bottom quintile of OScore (original paper shows non-monotonic returns, as does our replication)",1615.0
PatentsRD,Predictor,1_clear,2_fair,"Hirschleifer, Hsu and Li",2013,Patents to RD expenses,JFE,discrete,Other,profitability alt,1982,2008,PatentsRD,t=4.1 in FF3 style long-short,9A EMI1,LS FF3 style,1.0,0.41,4.13,VW,,,12.0,6.0,,"Table 9 does ff3 style VW to adjust for size.  Table 8 shows that predictability weak in large firms, so we just focus on small firms and VW to keep this spreadsheet manageable.","Drop obs before 1975.  Set expenses for research and development (xrd) and number of patents (npat)  to 0 if missing. Calculate Patents to RD as patents in a year divided by $xrd_{t-2} + .8 xrd_{t-3} + .6 xrd_{t-4} + .4xrd_{t-5} + .2 xrd_{t-6}$. Drop if denominator = 0,  2 years or less in Compustat, sic in 6000s, ceq < 0.  Double independent sort using (a) size NYSE median (b) terciles of Patents to RD.  Long if in small size, highest tercile of Patents to RD, short if in small size, smallest tercile.",1245.0
PayoutYield,Predictor,1_clear,2_fair,Boudoukh et al.,2007,Payout Yield,JF,continuous,Accounting,valuation,1984,2003,PayYield,monotonic NYSE decile sort but no LS t-stat,III B,port sort no LS,1.0,0.52,,EW,0.1,NYSE,12.0,6.0,,"Table III B NYSE deciles of positive yields; 0.52 is D10 minus D1, no t-stat in OP. Table VI B 30/70 sort (FF3 alpha t=3.92) is conservative and spec-sensitive; Table III is more complete. Our deciles track Table III closely. See issue #207.","Sum of dividends (dvc), purchase of common and preferred stock (prstkc) and max(pstkrv lagged 12 months minus pstkrv, 0), with an unavailable preferred-stock difference treated as zero, over market value of equity lagged 6 months.  Exclude if PayoutYield $\leq$ 0, financial firm based on SIC code, ceq <= 0, or less than 2 years in CRSP",777.0
PctAcc,Predictor,1_clear,1_good,"Hafzalla, Lundholm, Van Winkle",2011,Percent Operating Accruals,AR,continuous,Accounting,accruals,1989,2008,AccrOper,t>2.6 in size-adjusted long-short,4A,port sort size adjusted,-1.0,0.97,3.29,EW,0.1,,12.0,6.0,abs(prc)>5,"Table 4 Panel A: Report size-adjusted return hedges with p-values; t-stat is approximate, I converted the p-value to t but not exact since the table says p <.001 instead of giving a value.","Income before extraordinary items (ib) minus net cash flow (oancf) divided by absolute value of ib. If oancf is missing, PctAcc is defined as ( (act - act$_{t-12}$) - (che - che$_{t-12}$) - (  (lct - lct$_{t-12}$) - (dlc - dlc$_{t-12}$) - (txp - txp$_{t-12}$) - dp ) )/abs(ib). In either case, if ib is equal to 0, divide by .01 instead. Exclude if price less than 5.",211.0
PctTotAcc,Predictor,1_clear,1_good,"Hafzalla, Lundholm, Van Winkle",2011,Percent Total Accruals,AR,continuous,Accounting,accruals,1989,2008,AccrPct,t>2.6 in size-adjusted long-short,5A,port sort,-1.0,0.71,3.29,EW,0.1,,12.0,6.0,,"Table 5 Panel A: Report size-adjusted return hedges with p-values; t-stat is approximate, I converted the p-value to t but not exact since the table says p <.001 instead of giving a value.","Net income (ni) minus (purchase of common and preferred stock (prstkcc) minus sale of common and preferred stock (sstk) plus dividends (dvt), cash flow from operations (oancf), from financing (fincf) and investment (ivncf)). Scaled by absolute value of net income.",211.0
Price,Predictor,1_clear,1_good,Blume and Husic,1973,Price,JF,continuous,Price,other,1932,1971,Price,t=3 in regressions,2 column c,mv reg,-1.0,,2.9,EW,,,1.0,12.0,,,Log of absolute value of price (prc).,187.0
PriceDelayRsq,Predictor,1_clear,1_good,Hou and Moskowitz,2005,Price delay r square,RFS,continuous,Price,lead lag,1964,2001,PriceDelayRsq,t =3.4 in port sort char adj,2B one year daily,port sort char adjusted,1.0,0.31,3.4,EW,0.1,,12.0,7.0,,"Called D1 in paper.  Our primary delay measure, we use the daily stock level version.  OP focuses on complicated two-stage portfolio-based version","Each July regress stock return on day $t$ on market return in $t, t-1, \ldots, t-4$ using observations from July 1 of the previous year to June 30 of the current year.  Then regress again with no market return lags.  PriceDelay Rsq = 1 - [Rsq from second regression]/[Rsq from first regression]",1376.0
PS,Predictor,1_clear,2_fair,Piotroski,2000,Piotroski F-score,JAR,continuous,Accounting,composite accounting,1976,1996,Pscore,t=5.59 in port sort nonstandard data lag,3A,port sort,1.0,1.958333333,5.594,VW,0.1,,1.0,12.0,,"OP does not explicitly say if it is equal or value weighting, but we find more similar results using VW.  OP also compounds returns within a year, and rebalances 4 months after the end of fiscal year.  To approximate, we do portfolio period = 1",Sum of nine indicator variables which are: 1 if net income (ib) greater 0; 1 if net cash flow (oancf) greater 0; 1 if return on assets (ib/at) increased relative to previous year; 1 if net cash flow greater net income; 1 if long-term debt to assets (dltt/at) declined over the previous year;  if current assets to current liabilities (act/lct) increased over the previous year; 1 if ebit/sale (ebit = ib + txt + xint) increased over the previous year; 1 if revenue to assets increased over the previous year; 1 if shrout $\leq$ shrout last year. Include highest quintile of book-to-market only.  Exclude if missing any of the input variables.,1956.0
RD,Predictor,1_clear,1_good,"Chan, Lakonishok and Sougiannis",2001,R&D over market cap,JF,continuous,Accounting,R&D,1975,1995,RD,strong port sort,"4, first year",port sort no LS,1.0,0.8875,,EW,0.2,,12.0,6.0,,"Table 4 has portfolio returns, but no-tstats.  Table 6 does not show LS, but it has t-stat for high = 4.44.",R&D expense (xrd) over market value of equity.,2982.0
RDAbility,Predictor,1_clear,1_good,"Cohen, Diether and Malloy",2013,R&D ability,RFS,continuous,Accounting,other,1980,2009,RDAbility,t=2.6 in double sort,2A Spread,double sort,1.0,1.35,2.61,EW,0.2,,1.0,6.0,abs(prc)>5,2A has main double sort results.  Works both EW and VW.  Lots of supporting evidence too.,"Regress log of sales growth (sale over sale in previous year) on log of (1+ xrd/sale) in 5 bivariate regressions with (1+xrd/sale) lagged by $1, \ldots, 5$ years. Run regressions over previous 8 years and require at least 6 non-missing observations. Also require at least half of past research and development observations to be non-zero. RDAbility is the mean of the coefficients on the five lags of log(1+xrd/sale). Set to missing if firm is not in the highest tercile of xrd/sale in a year or if reseach and development expenses are non-positive.",543.0
RDcap,Predictor,1_clear,1_good,Li,2011,R&D capital-to-assets,RFS,continuous,Accounting,asset composition,1980,2007,RDcap,t=2.6 in long-short,7 RDCAP small size,LS port,1.0,0.69,2.64,VW,0.5,,12.0,6.0,,Table 7 shows that it only works in small firms with a hedge return t-stat of 2.64.   Works fine EW of course.,R&D capital to assets is the weighted sum of lagged R&D expenditures scaled by the assets (at). We replace xrd with 0 if xrd is missing and compute the numerator as $xrd_t + .8 xrd_{t-1}+ .6xrd_{t-2} + .4 xrd_{t-3} + .2 xrd_{t-4}$. Replace RDcap with missing before  1980 or if firm is in upper two thirds of market cap (shrout*abs(prc)) distribution in a month.,709.0
RDIPO,Predictor,1_clear,1_good,"Gou, Lev and Shi",2006,IPO and no R&D spending,JBFA,discrete,Event,R&D,1980,1995,RDIPO,t=2.68 in port sort FF3+Mom alpha,8 row 2,LS 4 factor alpha,-1.0,0.76,2.68,EW,,,1.0,6.0,,,Binary variable equal to 1 if R&D expense (xrd) = 0 and IndIPO = 1.  0 otherwise.,277.0
RDS,Predictor,1_clear,1_good,Landsman et al.,2011,Real dirty surplus,AR,continuous,Accounting,composite accounting,1976,2003,RDirtSurp,t=5.8 in port sort,4,port sort FF3 alpha,1.0,0.333333333,5.84,EW,0.333333333,,12.0,6.0,,Table 4 shows hedge return against FF3 with t-stat of almost 6. Tercile sorts. Table 4 pooled hedge return 1 year,"Define Dirty Surplus as annual change in marketable securities adjustment msa plus annual change in retained earnings adjustment (recta) + .65 times the annual change in min(Unrecognized prior service cost (pcupsu) - Pension additional minimum liability (paddml),0). Real dirty surplus is the annual change in book equity (ceq) minus dirty surplus minus (net income (ni) minus dividends preferred (dvp)) + dividends (divamt) - end-of-fiscal-year-stock-price (prcc\_f)*annual change in common shares outstanding (csho).",75.0
RealizedVol,Predictor,1_clear,1_good,Ang et al.,2006,Realized (Total) Volatility,JF,continuous,Price,volatility,1963,2000,IdioVol,t=2.9 in port sort,6A,port sort,-1.0,0.97,2.86,VW,0.2,,1.0,6.0,,,Standard deviation of residuals from CAPM regressions using the past month of daily data.  Value weighted,6340.0
Recomm_ShortInterest,Predictor,1_clear,4_lack_data,"Drake, Rees and Swanson",2011,Analyst Recommendations and Short-Interest,AR,discrete,Analyst,recommendation,1994,2006,Recomm_ShortInterest,t=4.09 FF+Mom alpha in port sort,7b,port sort FF+Mom alpha,1.0,1.11,4.09,EW,,,1.0,1.0,,,Go long firms in lowest quintile of short interest (shortint/shrout) and lowest quintile of analyst recommendations (monthly consensus recommendation using the most recent analyst recommendation in the past 12 months). Go short firms in highest quintile of short interest and highest quintile of analyst recommendations.,276.0
ResidualMomentum,Predictor,1_clear,1_good,"Blitz, Huij and Martens",2011,Momentum based on FF3 residuals,JEmpFin,continuous,Price,momentum,1930,2009,MomResid,t=8 in long-short ff3+ alpha,2B 1M,LS FF+ alpha,1.0,0.933333333,8.218550486,EW,0.1,,1.0,12.0,abs(prc)>1,"1M in the table refers to the holding period, not the signal measurement period.","Run a rolling regression over 36 months of excess return (retrf) on excess market return (mktrf), size and value factors (smb, hml) and compute idiosyncratic returns as the one-month lagged residual. ResidualMomentum is the rolling mean of the residual divided by the rolling standard deviation of the residual, both computed over the past 11 months.",346.0
retConglomerate,Predictor,1_clear,1_good,Cohen and Lou,2012,Conglomerate return,JFE,continuous,Price,lead lag,1977,2009,RetConglomerate,t=5.5 in port sort,2B,port sort,1.0,1.18,5.51,EW,0.1,,1.0,6.0,abs(prc)>5,Also works VW (t=3.2),"Identify conglomerate firms as those with multiple OPSEG or BUSSEG entries in the Compustat segment data (and require that at least 80% of firm's total assets are covered by segment data). Compute monthly stock return at the 2-digit SIC level for stand-alone (non-conglomerate) firms only, and match those returns to conglomerates' segments. Compute weighted conglomerate return as the industry return of stand-alone companies, weighted with a conglomerate's total sales in each industry.",666.0
ReturnSkew,Predictor,1_clear,1_good,"Bali, Engle and Murray",2015,Return skewness,Book,continuous,Price,risk,1963,2012,RetSkew,t=4 in port sort,14.10 total,port sort,-1.0,0.47,4.01,EW,0.2,,1.0,6.0,,"OP finds negative and significant relationship EW, and positive and significant relationship VW.  They conclude that results are ""difficult to interpret.""",Skewness of daily returns (ret) over previous month.,444.0
ReturnSkew3F,Predictor,1_clear,1_good,"Bali, Engle and Murray",2015,Idiosyncratic skewness (3F model),Book,continuous,Price,risk,1963,2012,RetSkew3F,t=4.4 in port sort,14.10 total,port sort,-1.0,0.45,4.35,EW,0.2,,1.0,6.0,,"OP finds negative and significant relationship EW, and positive and significant relationship VW.  They conclude that results are ""difficult to interpret.""","Skewness of idiosyncratic returns computed as residuals from regression of daily excess returns (ret - rf) on Fama-French factors (mktrf, smb, hml) over the previous month. We require at least 15 non-missing observations.",444.0
REV6,Predictor,1_clear,1_good,"Chan, Jegadeesh and Lakonishok",1996,Earnings forecast revisions,JF,continuous,Analyst,earnings forecast,1977,1992,EPSrevise,t=4.1 in regression,7 R6,mv reg,1.0,,4.07,EW,,,1.0,12.0,,"Table 5 has portfolio returns, but no-tstats.  Tab 7 has large t-stats in regressions.   Tab5 has monthly rebalancing, but Tab 7's regressiosn use 6-month holding period.  We find only monthly rebalancing works in port sorts.","Define revisions as the change in the mean earnings estimate (meanest) for the next quarter from month t-1 to t, scaled by stock price in month t-1. REV6 is the sum of that variable from months t-6 to t.",3322.0
RevenueSurprise,Predictor,1_clear,2_fair,Jegadeesh and Livnat,2006,Revenue Surprise,JAE,continuous,Accounting,sales growth,1987,2003,RevSurprise,t>2.6 in many event studies,7 Model 1 SURGE,event study regression 6 months,1.0,,,EW,0.2,,1.0,6.0,,"No t-stats in many event studies, but many 1% significant results (e.g. table 6)",Define revenue per share as quarterly revenue (revtq) divided by quarterly common shares outstanding (cshprq). RevenueSurprise is the 4-quarter change in revenue per share minus the average 4-quarter change in revenue per share over the previous 2 years. RevenueSurprise is scaled by its standard deviation over the previous 2 years. Exclude if price less than 5.,506.0
RIO_Disp,Predictor,1_clear,1_good,Nagel,2005,Inst Own and Forecast Dispersion,JFE,discrete,13F,short sale constraints,1980,2003,RIO_Disp,t = 2.47 in conditional sort,2,port sort,1.0,0.54,2.47,EW,,,1.0,12.0,,"We deviate a bit to avoid getting into IBES detail file.  Since we only use the summary file, we screen stdev > 0 and also keep both 4th and 5th quintiles of dispersion.","Follow RIO\_MB, except define Disp as the stdev of IBES forecasts where fpi == 1 divided by at, and sort on Disp instead of MB.  Finally, let RIO\_Disp = lagged RIO quntile if the Disp quintile >= 4.",1500.0
RIO_MB,Predictor,1_clear,1_good,Nagel,2005,Inst Own and Market to Book,JFE,discrete,13F,short sale constraints,1980,2003,RIO_MB,t = 4.91 in conditional sort,2B,port sort,1.0,1.07,4.91,EW,,,1.0,12.0,,,"Residual institutional ownership (RIO) is defined as log(institutional ownership (instown\_perc)/(1-institutional ownership)) + 23.6 - 2.89*log(market value of equity) +.08*log(market value of equity)$^2$. Replace instown\_perc with 0 if it is missing, with .9999 if it's above .9999, and with .0001 if it's below .0001. Then lag RIO by 6 months.  Keep only stocks with size > 20th percentile across exchcd == 1 or 2.  Define MB as market cap / (ceq + txditc), where txditc = 0 if missing, and MB is replaced with missing if ceq < 0.  Then sort into quintiles based on the lagged RIO, and sort into quintiles based on MB.  Finally, let RIO\_MB = lagged RIO quintile if the MB quintile = 5.",1500.0
RIO_Turnover,Predictor,1_clear,1_good,Nagel,2005,Inst Own and Turnover,JFE,discrete,13F,short sale constraints,1980,2003,RIO_Turnover,t = 2.71 in conditional sort,2,port sort,1.0,0.92,2.71,EW,,,1.0,12.0,,,"Follow RIO\_MB, except define turnover as vol/shrout.  Let RIO\_Turnover = lagged RIO quintile if the Volatility quintile == 5",1500.0
RIO_Volatility,Predictor,1_clear,1_good,Nagel,2005,Inst Own and Idio Vol,JFE,discrete,13F,short sale constraints,1980,2003,RIO_IdioRisk,t = 4.38 in conditional sort,2E,port sort,1.0,1.07,4.38,EW,,,1.0,12.0,,,"Follow RIO\_MB, except define volatility as the rolling standard deviation of the last 12 months of the stock return.  Let RIO\_Disp = lagged RIO quintile if the Volatility quintile == 5",1500.0
RIVolSpread,Predictor,1_clear,1_good,Bali and Hovakimian,2009,Realized minus Implied Vol,MS,continuous,Options,optionrisk,1996,2004,VRPSpread,t=3 in port sort,Table 3 Panel A,port sort,-1.0,0.732,2.9,VW,0.2,,1.0,6.0,"shrcd%in%c(10,11)",OP says NYSE only but we find all stocks gets closer to their numbers,Realized Vol (past 30 days) minus ATM Implied Vol,500.0
roaq,Predictor,1_clear,1_good,"Balakrishnan, Bartov and Faurel",2010,Return on assets (qtrly),JAE,continuous,Accounting,profitability,1976,2005,RoAq,"t=6.5 in port sort, nontraditional",Fig 1 Overall SAR,event study at portfolio-year level,1.0,0.841666667,6.45,EW,0.1,,1.0,12.0,abs(prc)>1,"This is like a more timely version of the other profitability measures.  Interestingly, they don't cite Fama French 2006, nor Novy Marx 2013.  MP have a very slightly different formulation.",Quarterly net income (ibq) divided by lagged total assets (atq).,338.0
sfe,Predictor,1_clear,2_fair,"Elgers, Lo and Pfeiffer",2001,Earnings Forecast to price,AR,continuous,Analyst,valuation,1982,1998,EPforecast,t=5 in long-short size adjusted,4A FAF_t+1 lower converage,LS size adjusted,1.0,1.466666667,4.99,EW,0.1,,12.0,3.0,abs(prc)>1,Size adjustment and analyst coverage are both extremely important. You can see the size adjustment in Table 2.,"Median estimate for next year eps (fpi = 1) in March, divided by stock price from December.  Dec fiscal year ends only, keep only forecasts more than 90 days out.  Keep only below median analyst coverage each month.",293.0
ShareIss1Y,Predictor,1_clear,1_good,Pontiff and Woodgate,2008,Share issuance (1 year),JF,continuous,Accounting,external financing,1970,2003,ShareIs5,t=7.08 in univariate reg,3A ISSUE,univariate reg,-1.0,,7.08,EW,,,12.0,6.0,,,Growth in number of shares between t-18 and t-6. Number of shares is calculated as shrout/cfacshr to adjust for splits.,874.0
ShareIss5Y,Predictor,1_clear,2_fair,Daniel and Titman,2006,Share issuance (5 year),JF,continuous,Accounting,external financing,1968,2003,ShareIs1,t=4.4 in univar reg,"3 \iota(t-5,t)",univariate reg,-1.0,,4.39,EW,,,12.0,6.0,,"Called N_t/N_t-\tau in OP.  OP actually studies this indirectly, see Equation (9) and close by.  Closest thing studied is \iota, which also includes dividends.",5-year growth in number of shares. Number of shares is calculated as shrout/cfacshr to adjust for splits.,1642.0
ShareVol,Predictor,1_clear,1_good,"Datar, Naik and Radcliffe",1998,Share Volume,JFM,discrete,Trading,volume,1962,1991,VolumeShare,t=8.9 in univariate reg,2A Turnover,univariate reg,-1.0,,8.86,EW,,,12.0,6.0,exchcd==1,"OP does a linear regression, but the variable is extremely right skewed, with a mean around 5, stdev of 18, and median of 2.  To approximate a regression in our setup, we long / short depending on the signal's value rather than ranking.","Let tempvol = sum of monthly share trading volume (vol) over the previous three months, scaled by 3 times common shares outstanding (shrout).  Let ShareVol = 1 if tempvol > 10%, and ShareVol = 0 if tempvol < 5%.",2136.0
Size,Predictor,1_clear,1_good,Banz,1981,Size,JFE,continuous,Price,size,1926,1975,Size,t=3.1 in long-short,3 \alpha_1 n=50,LS port CAPM alpha,-1.0,1.01,3.07,EW,0.5,,12.0,6.0,,,Log of monthly market value of equity (abs(prc)*shrout)).,11382.0
SmileSlope,Predictor,1_clear,1_good,Yan,2011,Put volatility minus call volatility,JFE,continuous,Options,optionrisk,1996,2005,OSmirkCP,t=8 in port sort,5A,port sort,-1.0,1.8,8.168,EW,0.2,,1.0,12.0,,,"Using OptionMetrics's daily volatility surfaces (vsurfd), keep last observation each month, delta = 0.50 or -0.50, and days to expiration = 30.  The signal is then the difference between put implied vol and call implied vol.",417.0
std_turn,Predictor,1_clear,1_good,"Chordia, Subra, Anshuman",2001,Share turnover volatility,JFE,continuous,Trading,liquidity,1966,1995,TurnovVol,t=3.7 in regression,5B TURN,mv reg,-1.0,,3.74,EW,,,1.0,6.0,exchcd==1,CVTURN and TURN in OP. Tab 3B has port sort but no LS or t-stats.  Tab 5B has FM reg.,Standard deviation of turnover (vol/shrout) over the past 36 months.,1349.0
STreversal,Predictor,1_clear,1_good,Jegadeesh,1990,Short term reversal,JF,continuous,Price,short-term reversal,1934,1987,Mom1m,t=12 in port sort,"2, S1, Jan-Dec",port sort CAPM alpha,-1.0,1.99,12.44,EW,0.1,,1.0,6.0,,,Stock return (ret) over the previous month.,4115.0
SurpriseRD,Predictor,1_clear,1_good,"Eberhart, Maxwell and Siddique",2004,Unexpected R&D increase,JF,discrete,Accounting,R&D,1974,2001,SurpriseRD,t=3.5 in long-short,5A EW,long port FF3 alpha,1.0,0.294,3.540083799,EW,,,1.0,6.0,,"Table 3 has event study, table 5 has LS ports.  Table 5 p-value = 0.000, so we use norm dist assuming p-value = 0.0004.  FF3 loadings should roughly cancel out.","Binary variable equal to 1 if: R&D (xrd) scaled by revenue (revt) is positive, R&D scaled by total assets (at) is positive, annual R&D growth is greater than 5%, annual growth in R&D over total assets is greater than 5%. SurpriseRD is 0 otherwise.",1408.0
tang,Predictor,1_clear,1_good,Hahn and Lee,2009,Tangibility,JF,continuous,Accounting,asset composition,1973,2001,Tangibility,t=3.37 in univariate FMB,4A Constrained,univariate reg,1.0,,3.37,EW,,,1.0,6.0,,"Paper is very painful to read.  Defines tangibility using formula we follow and used by Almeida and Campello.  Measures financial constraints using either asset size, payout ratio, etc, with bottom 30% classified as constrained.  Table 4 finds that tangibility only predicts returns in constrained firms in fm reg.  Table 7 forms FF3-style LS ports using only constrained firms and finds mixed results.  We follow Table 4.","Cash and short-term investments (che) plus .715*receivables (rect) + .547*inventory (invt) + .535* property, plant and equipment (ppent), scaled by total assets (at). Only defined for manufacturing firms (SIC $\geq$ 2000 and SIC <4000). Exclude the lowest tercile of manufacturing firms by total assets.",243.0
Tax,Predictor,1_clear,1_good,Lev and Nissim,2004,Taxable income to income,AR,continuous,Accounting,other,1973,2000,Tax2E,t=3.9 in regression,5A R_TAX,mv reg,1.0,,3.851,EW,,,12.0,6.0,abs(prc)>5,"Table 5 shows regressions, but it only works in the subsample 1973-1992.  Then again Table 6 shows 1993-2000 works as long as you drop 1998.  Focuses on forecasting earnings, like other accounting papers.","Ratio of Taxes paid and tax share of net income. Numerator is defined as the sum of foreign (txfo) and federal (txfed) income taxes. If either one is missing, numerator is defined as total taxes (txt) minus deferred taxes (txdi). Denominator is the product of the prevailing tax rate and net income (ib). Tax rate is .48 before 1979, .46 from 1979 to 1986, .4 in 1987, .34 between 1988 and 1992 and .35 from 1993 onwards. If net income is negative, and the numerator is positive, tax is defined as 1. Exclude if price less than 5.",871.0
TotalAccruals,Predictor,1_clear,1_good,Richardson et al.,2005,Total accruals,JAE,continuous,Accounting,investment alt,1962,2001,TotalAccruals,t=6 in mv reg,8A TACC,mv reg,-1.0,,6.38,EW,,,12.0,6.0,,Table 8 panel A has regression result with t-stat of 6.38.,"Before 1988: Change in net working capital ((act - che) - (lct - dlc)) plus change in net noncurrent assets ( (at - act - ivao) - (lt - dlc - dltt)) plus change in net financial assets ( (ivst + ivao - (dltt + dlc + pstk)). Starting in 1988: net income (ni) minus total, operating and investment cashflows (oancf, ivncf, fincf) plus stock sales minus repurchases and dividends (sstk, prstkc, dv)). Scaled by lagged total assets (at). Replace missings in ivao, ivst, dltt, dlc, pstk sstk, prstkc, dv with 0.",394.0
TrendFactor,Predictor,1_clear,1_good,"Han, Zhou, Zhu",2016,Trend Factor,JFE,continuous,Price,momentum,1930,2014,TrendFactor,t=15.0 in port sort,1,port sort,1.0,1.63,15.0,EW,0.2,,1.0,6.0,,"Filters (exchd in 1,2,3; sharecd in 10, 11; abs(prc)>5; mve>lowest decile of NYSE breakpoint distribution) are imposed at the signal stage in order to run cross-sectional regressions on the correct sample",See paper section 2.1 and 2.2,208.0
VolSD,Predictor,1_clear,1_good,"Chordia, Subra, Anshuman",2001,Volume Variance,JFE,continuous,Trading,liquidity,1966,1995,VolumeSD,t=3.6 in regression,5B DVOL,mv reg,-1.0,,3.56,EW,,,1.0,6.0,exchcd==1,CVVOL and VOL in OP.  Tab 3B has port sort but no LS or t-stats.  Tab 5B has FM reg,Rolling standard deviation of monthly trading volume (vol) over the past 36 months (require at least 24 observations). Include only NYSE stocks.,1349.0
XFIN,Predictor,1_clear,1_good,"Bradshaw, Richardson, Sloan",2006,Net external financing,JAE,continuous,Accounting,external financing,1971,2000,ExtFinNet,t=5.7 in port sort,3,port sort size adjusted,-1.0,1.29,5.7,EW,0.1,,12.0,6.0,,,Sale of common stock (sstk) minus dividends (dv) minus purchase of common stock (prstkc) plus long-term debt issuance (dltis) minus long-term debt reductions (dltr). Scaled by total assets (at).,798.0
zerotrade6M,Predictor,1_clear,1_good,Liu,2006,Days with zero trades,JFE,continuous,Trading,liquidity,1960,2003,ZeroTrade,t=4.1 in port sort,"2 LM6, Hp1m",port sort,1.0,0.806,4.06,EW,0.1,,1.0,12.0,exchcd==1,"Table 2, fourth row (LM6, HP1m), column B. Paper has the fancy title ""the LAPM""  Also works for longer holding period of up to 2 years.","In each month, count the number of days with no trades. Define zerotrade as the number of days without trades plus (the sum of monthly turnover (vol/shrout) divided by 48*10$^5$), multiplied by 21/number of trading days per month. Zerotrade is the 6-month average of that variable.",1521.0
zerotrade1M,Predictor,1_clear,1_good,Liu,2006,Days with zero trades,JFE,continuous,Trading,liquidity,1960,2003,zerotradeAlt1,t = 3.46 in port sort (12m holding),"2 LM1, Hp12m",port sort,1.0,0.56,3.46,EW,0.1,,12.0,12.0,exchcd==1,"The 1-month version works for 6 or 12 month holding periods, but not 1 month.","In each month, count the number of days with no trades. Define zerotrade as the number of days without trades plus (the sum of monthly turnover (vol/shrout) divided by 48*10$^5$), multiplied by 21/number of trading days per month. Take 1-month average.",1521.0
zerotrade12M,Predictor,1_clear,1_good,Liu,2006,Days with zero trades,JFE,continuous,Trading,liquidity,1960,2003,zerotradeAlt12,t > 4 in port sort (diff holding periods),"2 LM12, Hp1m",port sort,1.0,0.846,4.4,EW,0.1,,1.0,12.0,exchcd==1,Also works for 12 or even 24 months.,"In each month, count the number of days with no trades. Define zerotrade as the number of days without trades plus (the sum of monthly turnover (vol/shrout) divided by 48*10$^5$), multiplied by 21/number of trading days per month. Take 12-month average.",1521.0
Activism2,Predictor,2_likely,1_good,Cremers and Nair,2005,Active shareholders,JF,continuous,13F,ownership,1990,2001,Activism2,t=2.0 in port sort,4A VW EXT=4,port sort CAPM alpha,1.0,0.661666667,2.0,VW,0.25,,1.0,6.0,,works about the same EW in Tab 3,"Institiutional ownership share (maxinstown\_perc) if share greater than 5 percent.  Set to missing if G is missing, or if dual share class, or if 24 minus Governance index (G) is less than 19.",2037.0
AdExp,Predictor,2_likely,1_good,"Chan, Lakonishok and Sougiannis",2001,Advertising Expense,JF,continuous,Accounting,R&D,1975,1996,AdExp,53 bps spread but no t-stat,"7, first year",port sort no LS,1.0,0.529166667,,EW,0.2,,12.0,6.0,,"Most of the paper is about R&D, AdExp is kinda an afterthought.",Advertising expense (xad) over market value of equity (shrout*abs(prc)),2972.0
AgeIPO,Predictor,2_likely,2_fair,Ritter,1991,IPO and age,JF,continuous,Event,other,1975,1987,AgeIPO,"Event study, no t-stat",9A,event study 3 year,1.0,0.972222222,,EW,0.2,,1.0,6.0,,"Tab IX event study, no t-stat, but the magnitudes are crazy. Table 9, matching firm adjusted returns three year.  We follow MP.  Hand returns use 40 for long port and 5 for short port 36 month return for simplicity.  OP uses 1,500 IPOs total, so this cut would only use about 600 events, which is borderline.",Age is  (current year -  founding year from Jay Ritter's dataset). Exclude if IndIPO == 0.,6664.0
AnalystValue,Predictor,2_likely,2_fair,Frankel and Lee,1998,Analyst Value,JAE,continuous,Analyst,valuation,1975,1993,AnalystValue,p<0.01 in port sort but nonstandard stats,3D Ret12,port sort nonstandard p-val,1.0,0.258333333,,EW,0.2,,12.0,6.0,abs(prc) > 1,"Usually called V_f/P or V_f in OP.  OP finds p-val < 0.01, but spread is only 26 bps and sample is only 15 years long.  Also, p-value comes ""is derived using Monte Carlo simulation.  Specifically, we form empirical reference distributions by randomly assigning eligible firms into quintiles each year.""  Not sure OP's random assignment turned out right, since standard p-values are much higher than theirs for their Book-to-Price portfolios, which should be easy to replicate.  In the end, we judge predictability in OP as likely, since the nonstandard p-value, small return spread, and short sample suggest our test may only be borderline.","Value based on a three-stage dividend discount model and analyst forecasts, scaled by market value.  See code for details.",1739.0
AOP,Predictor,2_likely,2_fair,Frankel and Lee,1998,Analyst Optimism,JAE,continuous,Analyst,other,1975,1993,AOP,p<0.01 in port sort but nonstandard stats,5C OP Ret36,port sort nonstandard p-val,-1.0,0.275,,EW,0.2,,12.0,6.0,abs(prc) > 1,Called OP (optimism) in paper.   See AnalystValue.,"AnalystValue (defined above) minus IntrinsicValue (defined above), divided by abs(IntrinsicValue).",1739.0
Beta,Predictor,2_likely,1_good,Fama and MacBeth,1973,CAPM beta,JPE,continuous,Price,risk,1929,1968,Beta,t=2.6 univar reg,3A t(gamma_1),univariate reg,1.0,,2.57,EW,,,1.0,6.0,,"""and then risk-return regressions of (10) are fit month by month"" on page 617",Coefficient of a 60-month rolling window regression of monthly stock returns minus the riskfree rate on market return minus the risk free rate (ewretd - rf). Exclude if estimate based on less than 20 months of returns.,21173.0
BetaFP,Predictor,2_likely,3_distant,Frazzini and Pedersen,2014,Frazzini-Pedersen Beta,JFE,continuous,Price,other,1929,2012,BetaFP,t=7 in nonstandard port sort,3 BAB,port sort nonstandard,1.0,0.7,7.12,EW,0.1,,1.0,6.0,,"Though our signal should be close to the original, our portfolios are not.  The paper shows decile portfolios (table 3) but the factor is actually not constructed from going long and short the extreme deciles. Instead, the factor return is based on a median split on BetaFP, and then weighting within each portfolio by the rank of a stock's BetaFP (higher weight for high rank in above median portfolio, and higher weight for low rank in below median portfolio). They also standardize portfolios to have a beta of 1.","Using daily data, call tempRi the sum of today's return and the past 2 days, call tempRm the sum of today's mkt return and the past 2 days mkt return.  Regress return on tempRi and tempRm using past 5 years of data, with a minimum of 3 years.  BetaFP is the square root of the r square from this regression times the ratio of the idiosyncratic stock vol and the market vol.",323.0
BidAskSpread,Predictor,2_likely,2_fair,Amihud and Mendelson,1986,Bid-ask spread,JFE,continuous,Trading,liquidity,1961,1980,BidAskSpread,strong port sorts but no LS special data,2 Spread Mean,double sort no LS,1.0,,,EW,0.142857143,,1.0,1.0,,"Table 4 looks to be the closest; TZ: Table 2 gives portfolio returns, need to take difference between extreme portfolios ((7) - (1)) but no t-stat.  We use Corwin Schulz spread following MP, but OP uses Fitch's Stock Quotations on the NYSE.",Effective bid ask spread based on Corwin-Schulz scaled by stock price.,9029.0
BrandInvest,Predictor,2_likely,1_good,"Belo, Lin and Vitorino",2014,Brand capital investment,RED,continuous,Accounting,investment alt,1975,2010,BrandInvest,t=2.0 in port sort,1A r^S,port sort,-1.0,0.435,2.01,EW,0.2,,12.0,6.0,,OP's stock weighting is VW with a cap on the max weight for a single stock being 10%.  We do EW for simplicity.,"Advertising expenses (xad) divided by BrandCapital. Brand capital is computed by the perpetual inventory method. In the first year, brand capital is advertising expense divided by (.5 + .1). In subsequent years, we let brand capital depreciate with a rate of .5 and add current advertising expenses. Brand capital is scaled by total assets (at). Set to missing if advertising expense is missing.  Drop if 1st digit of sic = 4 or 6, and keep only Dec fiscal year ends.",283.0
ConsRecomm,Predictor,2_likely,4_lack_data,Barber et al.,2001,Consensus Recommendation,JF,discrete,Analyst,recommendation,1985,1997,ConsRecomm,t=3.2 in port sort nonstandard data,3A,port sort nonstandard data,-1.0,0.79,3.197,VW,0.2,,1.0,6.0,,"Main results use daily rebalancing, so we use Table 6C . Portfolio defs are in Tab IV caption.  Our data only begins in 1993, so it's impossible for us to replicate their results (their sample is 1986-1996).  However, we get similar results to theirs for the 1993-2003 sample.","Binary variable if the monthly mean of recommendations (ireccd) over analysts is greater than 3, and 0 if it is less or equal than 1.5.",1859.0
Coskewness,Predictor,2_likely,1_good,Harvey and Siddique,2000,Coskewness,JF,continuous,Price,risk,1964,1993,Coskew,p-val<0.05 in long-short,in text p 1276,LS port,-1.0,0.3,1.96,VW,0.3,,1.0,12.0,shrcd<=11,"The text reports 3.60 percent annual hedge return and p value < 0.05 (page 1278), but no table or further details.  OP states CAPM residuals are used, but we find we replicate OP very closely using simple de-meaning, which follow ACX.  ACX report private conversations with Harvey about replicating Coskewness in their paper.","Signal is the sample counterpart of $E[\tilde{r}_{it} \tilde{r}_{mt}^2]/( SD[\tilde{r}_{it} ] SD[\tilde{r}_{mt}]^2$ where $\tilde{r}_{it}$ is the de-meaned stock excess return and $\tilde{r}_{mt}$ is the de-meaned market excess return.  Signal is computed using the past 60 months of monthly data, and using the NYSE/AMEX CRSP VW index for the market (msic).  See code for details.",3704.0
DebtIssuance,Predictor,2_likely,2_fair,Spiess and Affleck-Graves,1999,Debt Issuance,JFE,discrete,Accounting,external financing,1975,1989,DebtIssuance,t = 2.19 FF3 alpha on long port,4A EW,long less risk-free ff3 alpha,-1.0,0.29,2.19,EW,,,1.0,6.0,,"OP uses ""Investment Dealers' Digest Directory of Corporate Financing"" but we use Compustat.",Equal to 1 if debt issuance (dltis) greater 0 and 0 otherwise. Exclude if share code > 11 or missing book-to-market.,589.0
DelDRC,Predictor,2_likely,1_good,Prakash and Sinha,2013,Deferred Revenue,CAR,continuous,Accounting,investment alt,2002,2007,DeferRev,t=3.6 in nonstandard reg  5 year sample,7 Delta DRC,univariate reg,1.0,,3.59,EW,,,1.0,6.0,,"OP shows only cross-sectional regressions, and the sample is very small (5 years). Table 7 has return forecasting regressiosn but no hedge returns. Other tables examine analyst forecast errors.","Annual change in deferred revenue (drc) scaled by average total assets (at) in t-1 and t. Exclude if negative book equity (ceq), deferred revenue equal to 0 in both years, revenue less than 5m, or SIC code between 6000 and 6999.",77.0
DownRecomm,Predictor,2_likely,4_lack_data,Barber et al.,2001,Down forecast EPS,JF,discrete,Analyst,earnings forecast,1985,1997,DownRecomm,t>8 in 3-day event study,"3C, 2 to 1, 3 to 2, etc.",event study 3 day nonstandard data,-1.0,,,EW,,,1.0,12.0,,"We follow MP and measure changes in earnings forecasts, while OP studies changes in recommendation.","Keep fpi = 1.  Binary variable equal to 1 if mean analyst earnings forecast for the next quarter (meanest) has improved over the previous month, and 0 otherwise.",1859.0
FirmAge,Predictor,2_likely,4_lack_data,Barry and Brown,1984,Firm age based on CRSP,JFE,continuous,Other,info proxy,1931,1980,FirmAge,t=2.5 in reg nonstandard data,"3 full samp, period of listing",mv reg nontstandard data,-1.0,,2.48,EW,,,1.0,6.0,exchcd==1,OP uses special NYSE archive data that we lack.,Months since start of CRSP coverage.,949.0
FR,Predictor,2_likely,2_fair,Franzoni and Marin,2006,Pension Funding Status,JF,continuous,Accounting,composite accounting,1980,2002,PensionFunding,49 bps long-short,3B,port sort no LS,1.0,0.4875,,EW,0.1,NYSE,12.0,6.0,,"Table 2 has port sorts but no LS t-stat.  Returns are non monotonic, but lowest FR seems to have clearly worst performance, EW or VW.  Table 3 has longer holding periods, and the pattern is more robust, but still no LS.  OP focuses on negative FR for technical reasons that we don't worry about.  VW is very weak, so we focus on EW.","FR = (FVPA - PBO), scaled by market value of equity. FVPA is pbnaa from 1980 to 1986, pplao + pplao from 1987 to 1997, and pplao after 1997. PBO is pbnvv from 1980 to 1986, pbpro + pbpru from 1987 to 1997, and pbpro after 1997. Exclude if price less than 5 or  shrcd > 11.",324.0
GrLTNOA,Predictor,2_likely,1_good,"Fairfield, Whisenant and Yohn",2003,Growth in long term operating assets,AR,continuous,Accounting,investment,1964,1993,LTNOAgr,61 bps spread in long-short,5A and B,port sort but no LS FF3 adjusted,1.0,0.61,,EW,0.1,,12.0,6.0,abs(prc)>5,"Long port has t=3.2 by itself, so it's a judgment call.","Annual growth in net operating assets, minus accruals. Net operating assets are (rect + invt + ppent + aco + intan + ao- ap- lco- lo) / at. Accruals are ( rect-l12.rect + invt - l12.invt + aco - l12.aco - (ap - l12.ap + lco - l12.lco) - dp ) / ((at + l12.at)/2)",1143.0
GrSaleToGrInv,Predictor,2_likely,1_good,Abarbanell and Bushee,1998,Sales growth over inventory growth,AR,continuous,Accounting,sales growth,1974,1988,RevG2InvG,t=2.4 in mv reg,2b RINV,reg,1.0,,2.372,EW,,,12.0,6.0,,,"Percentage growth in sales (sale) relative to average sales of t-1 and t-2, minus percentage growth in inventory (invt) relative to average inventory of t-1 and t-2. Both growth terms are calculated relative to t-1 only if t-2 is missing.",1123.0
GrSaleToGrOverhead,Predictor,2_likely,1_good,Abarbanell and Bushee,1998,Sales growth over overhead growth,AR,continuous,Accounting,sales growth,1974,1988,RevG2OHG,t=2.1 in mv reg,2b RS&A,reg,1.0,,2.069,EW,,,12.0,6.0,,"In our code, returns are only monotonic in quintiles 1-4, then they drop a lot in 5, perhaps consistent with OP's mv rank regressions.  In older versionsn we removed the 5th quintile, but now we keep it all in since OP did not hint at the nonmonotonicity.","GrSaleToGrOverHead = Percentage growth in sales (sale) relative to average sales of t-1 and t-2, minus percentage growth in administrative expenses (xsga) relative to average administrative expenses of t-1 and t-2. Both growth terms are calculated relative to t-1 only if t-2 is missing.",1123.0
HerfAsset,Predictor,2_likely,1_good,Hou and Robinson,2006,Industry concentration (assets),JF,continuous,Other,other,1963,2001,HerfAsset,t = 2.12 in characteristics-adjusted port sort,2 H(Assets),port sort char adjusted,-1.0,0.2,2.12,EW,0.2,,1.0,6.0,,"OP uses characteristic adjustments and gets a t-stat of 2.12 and 20 bps per month.  Not sure if raw return would exceed 1.96,","Three-year rolling average of the three digit industry Herfindahl index based on firm assets (at). Exclude regulated industries (4011, 4210, 4213 & year $\leq$ 1980; 4512 & year $\leq$ 1978, 4812, 4813 & year $\leq$ 1982, 4900-4999 in any year)",1316.0
High52,Predictor,2_likely,1_good,George and Hwang,2004,52 week high,JF,continuous,Price,momentum,1963,2001,High52,t=2.0 in long-short,1 High 52,port sort,1.0,0.45,2.0,EW,0.3,,6.0,6.0,,"Table 1 shows t=2.00 in LS.  Tab 2 shows sign depends entirely on January or non-January, and t-stats are huge in each subsample.  Should work really well VW, not sure why they don't do this.  Also works with price filter.",Price (abs(prc)/cfacpr) divided by the maximum price over the previous 12 months.,1177.0
IO_ShortInterest,Predictor,2_likely,1_good,Asquith Pathak and Ritter,2005,Inst own among high short interest,JFE,continuous,13F,ownership,1980,2002,InstOwnSI,strong port sort but no long-short,5 EW 99th,port sort 4-factor alpha difference,1.0,0.98,,EW,0.33,,1.0,12.0,exchcd==1,"Table 5 does not have long short returns in IO variable but shows returns for each tercile conditional on short interest. Subjectively large difference between lowest and highest IO terciles in alphas for EW returns, but the standard error can be large",Exclude all stocks with short interest (ShortInterest) below 99th percentile. IO\_ShortInterest is institutional ownership (instown\_perc).  Keep NYSE Only.,1302.0
MomOffSeason11YrPlus,Predictor,2_likely,1_good,Heston and Sadka,2008,Off season reversal years 11 to 15,JFE,continuous,Price,other,1965,2002,MomOffSeason11YrPlus,"t=1.8 in port sort, but similar strats do better",2 Years 11-15 annual,port sort,-1.0,0.19,1.77,EW,0.1,,1.0,6.0,"exchcd%in%c(1,2)","Considering the robustness of this signal, this ""small"" t-stat may be a fluke, and a Bayes-Stein or other multiple testing adjustment could actually increase the t-stat magnitude. But of course calling this a likely predictor is a judgment call.",Average return in other months over the preceding 11-15 years.,441.0
MRreversal,Predictor,2_likely,3_distant,De Bondt and Thaler,1985,Medium-run reversal,JF,continuous,Price,long term reversal,1933,1980,Mom1813,large ret in similar long-short,Fig2 two-year line,LS port nonstandard (plot only),-1.0,0.75,,EW,0.2,,12.0,6.0,,"We include this to nest MP.  Figure 2, two-year line is closest. Table I only shows LRreversal.  MP cite Jegadeesh and Titman 1993 JF but it's found in De Bondt and Thaler 1985 (or something close to it anyway),",Stock return between months t-18 and t-13.,13431.0
NumEarnIncrease,Predictor,2_likely,3_distant,Loh and Warachka,2012,Earnings streak length,MS,continuous,Accounting,earnings growth,1987,2009,NumEarnIncrease,similar results in port sorts but not exact,4 all together,port sort FF3 alpha,1.0,,,EW,0.2,,1.0,12.0,abs(prc)>5,"This signal isn't exactly found in Loh and Warachka, but table 4 suggests that this would predict returns.  GHZ have this signal and cite Barth, Elliott and Finn 1999, but Barth et al do not study predictability according to Loh and Warachka.  I (Andrew) couldn't locate Barth et al online during the pandemic, but the abstract of Barth et al is consistent with Loh and Warachka's interpretation.","Number if consecutive 4-quarter increases in ibq, up to 8.",92.0
OperProf,Predictor,2_likely,1_good,Fama and French,2006,operating profits / book equity,JFE,continuous,Accounting,profitability,1977,2003,ProfOper,t=2.6 in mv reg,3 Y_t/B_t,mv reg,1.0,,2.55,EW,,,12.0,6.0,,,"Revenue (revt) minus cost (cogs) - administrative expenses (xsga) - interest expenses (xint), scaled by book value of equity (ceq). Exclude smallest size tercile.",1599.0
OperProfRD,Predictor,2_likely,1_good,Ball et al.,2016,Operating profitability R&D adjusted,JFE,continuous,Accounting,profitability,1963,2014,OperProfRD,t=1.8 in port sort,4A Oper PRof,port sort,1.0,0.29,1.84,VW,0.1,NYSE,12.0,6.0,,"This is operating prof with an R&D adjustment.  Strong in FM reg with controls, but insignificant in VW port sorts (Tab 4). OP states they lag denominator, but 2015 JFE does not lag, no clear motivation to lag, and our replications much closer to OP without lag.","Revenue (revt) minus cost (cogs) - (administrative expenses (xsga) - R&D expenses (xrd)), all divided by total assets (at) in year t. Replace all variables in the numerator with 0 if they are missing. Exclude if share code is greater 11, market value of equity, BM or total assets are missing, or if SIC code between 6000 and 6999.",533.0
OptionVolume2,Predictor,2_likely,2_fair,Johnson and So,2012,Option volume to average,JFE,continuous,Options,volume,1996,2010,OptVolGr,t = 2.5 in port sort CAPM alpha weekly data,2A \Delta (O/S),port sort CAPM alpha nonstandard,-1.0,0.516,2.45,EW,0.2,,1.0,6.0,"abs(prc)>1, shrcd<=11","OP is actually weekly, and sorts in deciles, then  longs 9+10 and shorts 1+2",Based off of OptionVolume1.  OptionVolume2 = OptionVolume1 / average of OptionVolume1 from months t-6 to t-1.,457.0
PredictedFE,Predictor,2_likely,3_distant,Frankel and Lee,1998,Predicted Analyst forecast error,JAE,continuous,Accounting,earnings forecast,1979,1993,EPSforeErr,p<0.01 in reg but nonstandard stats,8A PErr,univariate reg nonstandard p-val,-1.0,,,EW,,,12.0,6.0,abs(prc) > 1,"Called PErr in OP.  Very difficult to understand what OP is doing, so we report something close in spirit which produced a consistent port sort (albeit statisticallly insignficiant).  See also AnalystValue.","The fitted value from cross-sectional regressions of analyst earnings' forecast errors on cross-sectional rankings of 5-year sales growth, book-to-market, AOP, and analyst long term growth.  See code for details.",1739.0
PriceDelaySlope,Predictor,2_likely,3_distant,Hou and Moskowitz,2005,Price delay coeff,RFS,continuous,Price,lead lag,1964,2001,PriceDelay,t =7.7 in port sort w/ complicated signal,2A D2 adjusted,port sort char adjusted,1.0,1.21,7.7,EW,0.1,,12.0,7.0,,"see PriceDelayRsq.  Called D2 in paper.  Related statistics shown in Table 2, but not our daily version.  Table 2 shows daily vs two-stage weekly makes a huge difference.","Each July regress daily stock return (ret) on market return (mktrf) in $t, t-1, \ldots, t-4$ with observations over the previous year. Define PriceDelaySlope as the ratio of ]1*beta on mktrf$_t-1$ + 2*beta on mktrf$_t-2$ + 3*beta on mktrf$_t-3$ + 4*beta on mktrf$_t-4$], and [beta on mktrf$_t$ + beta on mktrf$_t-1$ + beta on mktrf$_t-2$ + beta on mktrf$_t-3$ + beta on mktrf$_t-4$].",1370.0
PriceDelayTstat,Predictor,2_likely,3_distant,Hou and Moskowitz,2005,Price delay SE adjusted,RFS,continuous,Price,lead lag,1964,2001,PriceDelayAdj,t =7.39 in port sort w/ complicated signal,2A D3 adjusted,port sort char adjusted,1.0,1.1,7.39,EW,0.1,,12.0,7.0,,"see PriceDelayRsq.  Called D3 in paper.  Related statistics shown in Table 2, but not our daily version. Table 2 shows daily vs two-stage weekly makes a huge difference.","Each July regress daily stock return (ret) on market return (mktrf) in $t, t-1, \ldots, t-4$ using observations from July 1 of the previous year to June 30 of the current year.  Trim the highest and lowest 1% of estimated coefficients, standard errors, and t-stats. Define PriceDelayTstat as the ratio of [1*tstat on mktrf$_t-1$ + 2*tstat on mktrf$_t-2$ + 3*tstat on mktrf$_t-3$ + 4*tstat on mktrf$_t-4$], and [tstat on mktrf$_t$ + tstat on mktrf$_t-1$ + tstat on mktrf$_t-2$ + tstat on mktrf$_t-3$ + tstat on mktrf$_t-4$].   Finally winsor the extreme 10 and 90 percentiles each month.",1370.0
ProbInformedTrading,Predictor,2_likely,1_good,"Easley, Hvidkjaer and O'Hara",2002,Probability of Informed Trading,JF,continuous,Trading,liquidity,1984,1998,PIN,t=2.5 in mv reg,6,mv reg,1.0,,2.496,EW,0.2,,1.0,6.0,,"Table 3 has double sorts with size but no t-stats.  Sign of predictability depends on size, only increases for bottom 3 size quintiles.  Table 6 has mv reg, t=2.5 controlling for size, but size has t-stat of 2.8.","Download parameter estimates from the archived version of Hvidkjaers website (https://web.archive.org/web/20110219024112/http://sites.google.com/site/hvidkjaer/data/data-files/pin1983-2001.zip?attredirects=0), and the page for Duarte, Hu, Young, JFE 2020 https://edwinhu.github.io/pin/ and apply Easley et al's Equation 5. Use Hvidkjaers data when available, otherwise use Duarte et al. Use Year t PIN estimates to forecast returns throughout Year t+1",2835.0
realestate,Predictor,2_likely,1_good,Tuzel,2010,Real estate holdings,RFS,continuous,Accounting,asset composition,1971,2005,RealEstate,t=1.8 (VW) and t= 1.28 (EW) in port sort,5 r^e_VW,port sort,1.0,0.244166667,1.8,VW,0.2,,12.0,6.0,,OP finds smaller t-stat = 1.3 EW.  OP is mainly theory.  Does not describe data handling super closely,"Industry-adjusted value of real estate holdings. Real estate holdings are (fatb+fatl)/ppegt if available, and (ppenb+ppenl)/ppent otherwise.  Drop firms in 2 digit sics with < 5 firm or missing at or missing both ppent and ppegt. Subtract monthly industry-mean at the 2 digit SIC level.",249.0
RoE,Predictor,2_likely,2_fair,Haugen and Baker,1996,net income / book equity,JFE,continuous,Accounting,profitability,1979,1993,RoE,t=4.5 in mv reg nonstandard,1 return on equity,mv reg nonstandard,1.0,,4.5,EW,0.2,,12.0,6.0,abs(prc)>5,OP reports mean regression coeff across 90 multiple regressions.,Net income (ni) over book value of equity (ceq). Exclude if price less than 5.,1647.0
ShareRepurchase,Predictor,2_likely,3_distant,"Ikenberry, Lakonishok, Vermaelen",1995,Share repurchases,JFE,discrete,Accounting,payout indicator,1980,1990,ShareRepurchase,t=1.85 in long - benchmark port,3 All firms Year 1,long port char adjusted,1.0,0.17,1.852179859,EW,,,12.0,6.0,,,"Binary variable equal to 1 if stock repurchase indicated in cash flow statement (prstkc > 0), and 0 if prstkc = 0.",3055.0
ShortInterest,Predictor,2_likely,2_fair,Dechow et al.,2001,Short Interest,JFE,continuous,Trading,short sale constraints,1976,1993,ShortInterest,35 bps spread in port sort,1A overall ave,pooled mean,-1.0,0.35,,EW,0.2,,1.0,6.0,,"Table 1, Panel A has a low- and high- short abnormal return from pooled observations (not calendar time).  Standard errors are shown by group and range from 0.8 to 3.7 percent.  Hard to say since it's so far from our calendar time ports.  Most of the paper is about the correlation between short interest and valuations ratios.  Strange case where OP does not invesigate predictability much, but predictability is very strong.",Short-interest from Compustat (shortint) scaled by shares outstanding (shrout). Short-interest data are available bi-weekly with a four day lag. We use the mid-month observation to make sure data would be available in real time.,1117.0
sinAlgo,Predictor,2_likely,2_fair,Hong and Kacperczyk,2009,Sin Stock (selection criteria),JFE,discrete,Other,other,1926,2006,SinStock,t-stat = 1.8 in LS nontraditional,"4A 1965-2006, first row, 4B 1926-2006, first row",LS nonstandard,1.0,0.3,2.0,EW,,,1.0,6.0,,"Follows OP and uses Compustat segment data plus naics codes. SinAlgo is equal to 0 for ""comparable stocks"" (FF48 groups 2, 3, 7 and 43) ","Using Compustat Segment data, sinAlgo is defined as a binary variable equal to 1 if at least one segment of a firm is listed as being in at least one of the following industries: sic $\geq$ 2100 & sic $\leq$ 2199, sic $\geq$2080 & sic $\leq$ 2085, NAICS in \{7132, 71312, 713210, 71329, 713290, 72112, 721120\}. As in the original paper, we assume that the sin stock indicator applies to the entire history and future of the identified firm. sinAlgo is equal to 0 if the firm is not identified in the CS Segment data as a sin stock and if the firm is in one of the following industries: (sic $\geq$ 2000 & sic $\leq$ 2046) OR (sic $\geq$ 2050 & sic $\leq$ 2063) OR (sic $\geq$ 2070 & sic $\leq$ 2079) OR (sic $\geq$ 2090 & sic $\leq$ 2092) OR (sic $\geq$ 2095 & sic $\leq$ 2099) OR (sic $\geq$ 2064 & sic $\leq$ 2068) OR (sic $\geq$ 2086 & sic $\leq$ 2087) OR (sic $\geq$  920 & sic $\leq$  999) OR (sic $\geq$ 3650 & sic $\leq$ 3652) OR sic == 3732 OR  (sic $\geq$ 3931 & sic $\leq$ 3932) OR (sic $\geq$ 3940 & sic $\leq$ 3949) OR  (sic $\geq$ 7800 & sic $\leq$ 7833) OR (sic $\geq$ 7840 & sic $\leq$ 7841) OR (sic $\geq$ 7900 & sic $\leq$ 7911) OR (sic $\geq$ 7920 & sic $\leq$ 7933) OR (sic $\geq$ 7940 & sic $\leq$ 7949) OR sic $==$ 7980 OR (sic $\geq$ 7990 & sic $\leq$ 7999)",3655.0
skew1,Predictor,2_likely,2_fair,"Xing, Zhang and Zhao",2010,Volatility smirk near the money,JFQA,continuous,Options,optionrisk,1996,2005,OSmirkNTM,t = 2.19 in port sort,3A,port sort weekly,-1.0,0.64,2.19,EW,0.2,,1.0,12.0,,Tab3A shows weekly ret LS t-stat of 2.19.  Screens in appendix are important.,"Using OptionMetrics data, among options with duration between 10 and 60 days, implied volatility of put option with moneyness closest to but above 1 minus implied volatility of call option with moneyness closest to but below 1.  Keep only volume > 0, implied vol between 0.03 and 2.0, price > 0.125, open interest > 0.",945.0
SP,Predictor,2_likely,1_good,"Barbee, Mukherji and Raines",1996,Sales-to-price,FAJ,continuous,Accounting,valuation,1979,1991,Rev2Price,t=2.5 in mv reg,2 model 1,mv reg,1.0,,2.52,EW,,,12.0,6.0,,,Ratio of annual sales (sale) to market value of equity.,420.0
Spinoff,Predictor,2_likely,4_lack_data,"Cusatis, Miles and Woolridge",1993,Spinoffs,JFE,discrete,Event,other,1965,1988,Spinoff,t=2.3 in event study,3B I-24,event study nonstandard data,1.0,2.083333333,2.43,VW,,,1.0,12.0,,"OP uses CCH Capital Changes Reporter, but we use CRSP acquisition file.  More importantly OP excludes about 75% of spinoffs because they are not ""pure"" spinoffs, but we do not in order to keep a reasonable amount of stocks.  Tab 3B event study uses matched-firm-adjusted returns and holds for 2 years, for only 140 spinoffs.  Predictability is a judgment call, but if VW gets a t-stat of 2.43, it seems it should have a 50/50 shot.",Spinoffs are identified as all observations in the CRSP acquisition file with valid acperm entry. Spinoff is a binary variable equal to 1 if a firm is identified in the CRSP Acquisition data and if it has at most two years of history in the CRSP stock return data. Spinoff is equal to 0 otherwise.,617.0
UpRecomm,Predictor,2_likely,4_lack_data,Barber et al.,2001,Up Forecast,JF,discrete,Analyst,earnings forecast,1985,1997,UpRecomm,t>8 in 3-day event study,"3C, 1 to 2, 2 to 3, etc",event study 3 day nonstandard data,1.0,,,EW,,,1.0,12.0,,"We follow MP and measure changes in earnings forecasts, while OP studies changes in recommendation.",Keep fpi = 1.  Binary variable equal to 1 if mean earnings forecast (meanest) decreased over the past month.,1860.0
VarCF,Predictor,2_likely,2_fair,Haugen and Baker,1996,Cash-flow  to price variance,JFE,continuous,Accounting,cash flow risk,1979,1993,CF2Pvar,t=2.5 in mv reg nonstandard,1 variability in cf to price,mv reg nonstandard,-1.0,,2.5,EW,0.2,,12.0,6.0,,"OP reports mean regression coeff across 90 multiple regressions.  OP shows a minus sign in Tab 1, but we find monotonic returns with a plus sign, and the plus sign is consistent with traditional theory.",Rolling variance of (ib+dp)/mve\_c over the past 60 months (minimum 24 months data required).,1647.0
VolMkt,Predictor,2_likely,2_fair,Haugen and Baker,1996,Volume to market equity,JFE,continuous,Trading,volume,1979,1993,Volume2Mkt,t=4 in mv reg nonstandard,1 trading volume / market cap,mv reg nonstandard,-1.0,,4.0,EW,0.2,,12.0,6.0,abs(prc)>5,OP reports mean regression coeff across 90 multiple regressions.,"Average monthly dollar trading volume (vol*abs(prc)) over the previous 12 months, scaled by market value of equity. Exclude if price less than 5.",1647.0
VolumeTrend,Predictor,2_likely,2_fair,Haugen and Baker,1996,Volume Trend,JFE,continuous,Trading,volume,1979,1993,VolumeTrend,t=3 in mv reg nonstandard,1 trading volume trend,mv reg nonstandard,-1.0,,3.0,EW,0.2,,12.0,6.0,,OP reports mean regression coeff across 90 multiple regressions.,Rolling coefficient from regressing monthly trading volume on a linear time trend over a window of 60 months (require that at least 30 exist). Scale coefficient by 60-month average of trading volume.,1647.0
BetaBDLeverage,Placebo,4_not,1_good,"Adrian, Etula and Muir",2014,Broker-Dealer Leverage Beta,JF,continuous,Trading,liquidity,1973,2009,BetaBDLeverage,t=1 in conservative port sort,XIIB,port sort,1.0,0.3225,1.26,VW,0.2,,12.0,6.0,me > me_nyse10,"Strangely, the vol of our LS port is about 30% smaller than OP.  Also our returns are about 30% higher.  As a result, our t-stat is almost twice as large.",Regress quarterly stock return minus 3 month treasury bill rate (tbillrate3m) on broker dealer leverage. Use a rolling window of 40 quarters (require at least 20 non-missing observations). BetaBDLeverage is the coefficient on broker-dealer leverage. Broker-dealer leverage is the seasonally adjusted ratio of assets (FRED series BOGZ1FL664090005Q) and liabilities (FRED series BOGZ1FL664190005Q).,1030.0
BetaSquared,Placebo,4_not,1_good,Fama and MacBeth,1973,CAPM beta squred,JPE,continuous,Price,other,1929,1968,BetaSquared,t=0.3 in mv reg,3A t(gamma2),mv reg,-1.0,,0.29,EW,,,1.0,6.0,,,Square of Beta (defined above).,21186.0
BidAskTAQ,Placebo,4_not,1_good,Hou and Loh,2016,Bid-ask spread (TAQ),JFE,continuous,Trading,liquidity,1984,2012,BidAskTAQ,t=1.3 in mv reg,2 Model 10,mv reg,1.0,,1.26,EW,,,12.0,6.0,,"Insignificant with IVOL in mv reg.  Could maybe be significant without IVOL, so it's a judgement call.",TAQ-based trading costs estimates (SAS code).,375.0
DelayAcct,Placebo,4_not,1_good,"Callen, Khan and Lu",2013,Accounting component of price delay,CAR,continuous,Accounting,lead lag,1981,2006,DelayAcct,t=1 in long-short,7B,double sort FF3 alpha,1.0,0.28,1.08,EW,0.2,,12.0,6.0,,Tab 7b double sorts on the two delay components and generally finds insignificant LS ports.  We record the middle values for the first sort.  Variable is listed by HLZ,"Monthly cross-sectional regression of PriceDelay on AccrualQuality, special items (si) scaled by average total assets (at) and earnings surprise (meanest - actual) scaled by its cross-sectional standard deviation. DelayAcct is the predicted value from that regression.",411.0
DelayNonAcct,Placebo,4_not,1_good,"Callen, Khan and Lu",2013,Non-accounting component of price delay,CAR,continuous,Accounting,lead lag,1981,2006,DelayNonAcct,t=1 in long-short,7B,double sort FF3 alpha,1.0,0.41,1.11,EW,0.2,,12.0,6.0,,Tab 7b double sorts on the two delay components and generally finds insignificant LS ports.  We record the middle values for the first sort.  Variable is listed by HLZ,"Monthly cross-sectional regression of PriceDelay on AccrualQuality, special items (si) scaled by average total assets (at) and earnings surprise (meanest - actual) scaled by its cross-sectional standard deviation. DelayNonAcct is the residual  from that regression.",411.0
DelSTI,Placebo,4_not,1_good,Richardson et al.,2005,Change in short-term investment,JAE,continuous,Accounting,investment alt,1962,2001,DelSTI,t=0.4 in mv reg,8C DeltaSTI,univariate reg,-1.0,,0.43,EW,,,12.0,6.0,,,Difference of the current and one-year lagged short-term investments (ivst) scaled by the mean of current and one-year lagged total assets (at).,2297.0
DownsideBeta,Placebo,4_not,1_good,"Ang, Chen and Xing",2006,Downside beta,RFS,continuous,Price,risk,1963,2001,betaDown,t=0.6 in port sort,8A,port sort,1.0,0.11,0.6,EW,0.2,,1.0,6.0,exchcd==1,Tables 1-7 show only realized beta-.  Only on Table 8 do you see past beta- which is insignificant.,Beta of Daily stock return (ret - rf) regressed on market return (mktrf) for days for which the market return was less than the average market return over the previous year. Rolling window of 252 trading days with at least 50 observations.,1692.0
FailureProbability,Placebo,4_not,1_good,"Campbell, Hilscher and Szilagyi",2008,Failure probability,JF,continuous,Accounting,default risk,1981,2003,FailurePr,t=1.5 in conservative port sort,6A LS2080,port sort,-1.0,0.525,1.41,VW,0.2,,12.0,12.0,abs(prc)>1,"Though OP LS is insignificant, returns are mostly monotonic. OP's goal is not to show this is a predictor, but that it does not explain other predictors.","Failure probability is -9.16 -.058*PRICE + .075*MB - 2.13*CASHMTA -.045*RSIZE + 1.41*IdioRisk - 7.13*EXRETAVG + 1.42*TLMTA - 20.26*NIMTAAVG. PRICE is log(min(abs(prc), 15)); MB is shrout*abs(prc)/ceqq; CASHMTA is cheq/(shrout*abs(prc) + ltq); RSIZE is log(shrout*abs(prc)/ sum of shrout*abs(prc) for the largest 500 companies each month); IdioRisk is defined above, EXRETAVG is the weighted average excess return (log(1 + ret) - log(1 + mktrf)) over the previous 12 months, with weight on month t-j being $\phi^j$ and the sum scaled by $\frac{1-\phi}{1-\phi^{12}}$; TLMTA is total liabilities (ltq/(shrout*abs(prc)); NIMTAAVG is a weighted average of net income over total assets (ibq/(shrout*abs(prc) + ltq)) over four quarters, with weight $\phi^q$ on quarter $t-q$ and the sum scaled by $\frac{1-\phi^3}{1-\phi^{12}}$. $\phi = 2^{-\frac{1}{3}}$. All input variables are winsorized at the 5th and 95th percentile. Exclude if price less than 1.",3359.0
ForecastDispersionLT,Placebo,4_not,2_fair,"Anderson, Ghysels, and Juergens",2005,Long-term forecast dispersion,RFS,continuous,Analyst,volatility,1991,1997,EPSDispLT,t=1.0 in conservative long-short,2A,LS port,1.0,0.23,0.96,VW,0.5,,1.0,12.0,,"OP's focus is on pricing other assets, not the premium on this portfolio.  We did not study this super in depth.",Standard deviation of earnings estimates (stdev\_est) scaled by mean earnings estimate.  Keep if ME is in the top 500 for the month to approximate S&P 500.,341.0
KZ,Placebo,4_not,1_good,"Lamont, Polk and Saa-Requejo",2001,Kaplan Zingales index,RFS,continuous,Accounting,composite accounting,1968,1997,KZ,t=1.1 in conservative port sort,5 FC,LS port FF3 style,-1.0,0.13,1.102326615,VW,0.33,,12.0,6.0,,Most of the paper is about covariance analysis.  OP does VW with 3x3 FF3 style even though EW gets huge t-stats.  We just do VW.  Called FC in paper.  Table 5 has factor returns.,-1.002* (net income (ni) + depreciation (dp))/total assets (at) + .283*(total assets (at) + market value of equity - book value of equity (ceq) - deferred taxes (txdi))/total assets (at) + 3.319*(debt in current liabilities (dlc) + long-term debt (dltt))/(debt in current liabilities + long-term debt + book value of equity) - 39.368*(Dividends (divamt)/total assets) - 1.315*(cash and short-term investments (che)/total assets). Replace txdi and divamt with 0 if missing.,24.0
rd_sale,Placebo,4_not,1_good,"Chan, Lakonishok and Sougiannis",2001,R&D to sales,JF,continuous,Accounting,investment alt,1975,1995,rd_sale,8 bps spread in port sort,3A,port sort,1.0,0.08,,EW,0.2,,12.0,6.0,,"No t-stat in Tab 3, but authors say ""there is little if any relation between R&D relative to sales and future returns.""   Pattern is non monotinic, mostly increasing, but drops in port 5, so we have a negative sign",One year lagged R&D (xrd) divided by one year lagged sales (sale).,2974.0
roic,Placebo,4_not,1_good,Brown and Rowe,2007,Return on invested capital,WP,continuous,Accounting,profitability,1970,2005,ROIC,t=0.9 in port sort,1B,port sort,1.0,0.14,0.9,EW,0.2,,12.0,6.0,,,"EBIT (ebit) minus non-operating income (nopi) divided by the sum of equity (ceq), liabilities (lt) and cash (che).",92.0
WW,Placebo,4_not,2_fair,Whited and Wu,2006,Whited-Wu index,RFS,continuous,Accounting,external financing,1975,2001,WW,t=1.3 in port sort,4 FC Equal-Weighted,port sort ff3 style,1.0,0.23,1.32,EW,0.4,,12.0,12.0,,"OP shows ff3-style long-short portfolio with value-weighting first, but we focus on equal-weighting for simplicity.  Regardless the results are similar in OP, and OP states ""financially constrained firms earn a positive, albeit statistically insignificant risk premium.""",Group data by 3 digit SIC code and month to compute total sales (sale) by industry. Calculate the Whited-Wu index as -0.091*(income before extraordninary items (ib) + depreciation and amortizarion (dp)) / (4 * assets(at)) - 0.062( if the dividends per share (dvpsx_c) is not missing and greater than 0) + 0.021 * long-term debt (dltt) / at - 0.044 * log(at) + 0.102 * (twelve-month growth of Industry Sales) / 4 - 0.035 * (one-month growth of sales) / 4.,4286.0
ZScore,Placebo,4_not,1_good,Dichev,1998,Altman Z-Score,JF,continuous,Accounting,default risk,1981,1995,ZScore,t=1.59 in univar reg,3A,univariate reg,-1.0,,1.59,EW,,,1.0,6.0,exchcd==1,"Intro claims higher bankrupty prob has ""significantly"" lower returns, but Tab 3 shows the evidence is mixed for Z Score.  Z is only signficant in regs controlling for B/M, and has a nonmonotonic realtionship with ret.  Unlike for OSCore, there is no clear LS strategy.    Sign is wrong for nasdaq (Tab3B).  One could ""fix"" this signal as we did before by dropping the lowest ZScore quint, as hinted at in Tab 3A.","1.2*(current assets (act) - current liabilities (lct))/total assets (at) + 1.4*(Retained earnings (re)/total assets (at)) + 3.3*(net income (ni) + interest expense (xint) + total taxes (txt))/total assets (at) +  .6*(market value of equity/Total liabilities (lt)) + revenue (revt)/ total assets (at). Include only NYSE stocks. Exclude if SIC code between 4000 and 4999, or above 5999.  Exclude if ZScore is in bottom quintile of ZScore (original paper shows non-monotonic returns, as does our replication)",1611.0
ConsNegRet,Drop,9_drop,9_drop,Watkins,2003,Consistently negative return,Journal of Behavioral Finance,,Price,momentum,1927,1999,ConsNegRet,9_drop,9_drop,,1.0,,,EW,,,1.0,6.0,,"Not in either MP or HXZ and not top 3,",Define consistent negative returns (ConsNegRet) as 1 if stock returns from months t to t-5 are all less than 0.,2094.0
ConsPosRet,Drop,9_drop,9_drop,Watkins,2003,Consistently positive return,Journal of Behavioral Finance,,Price,momentum,1927,1999,ConsPosRet,9_drop,9_drop,,1.0,,,EW,,,1.0,6.0,,"Not in either MP or HXZ and not top 3,",Define consistent positive returns (ConsPosRet) as 1 if stock returns from months t to t-5 are all greater than 0.,2094.0
invest,Drop,9_drop,9_drop,Chen and Zhang,2010,Capex and Inventory Change,"JF, but retracted",continuous,Accounting,investment,1972,2006,invest,drop,9_drop,,-1.0,,,EW,0.2,,1.0,6.0,,"Should remove.  Paper was retracted,","Annual change in property, plant and equipment (ppegt) plus annual change in inventory (invt), scaled by lagged total assets (at). Use ppent if ppegt is missing.",250.0
PosNegCons,Drop,9_drop,9_drop,Watkins,2003,Pos vs negative consistent return,Journal of Behavioral Finance,,Price,momentum,1950,1999,PosNegCons,9_drop,9_drop,,1.0,,,EW,,,12.0,6.0,,"Not in either MP or HXZ and not top 3,","Binary signal equal to if ConsPosRet equal to 1, and equal to 0 if CosnNegRet equal to 1.",46.0
sinOrig,Drop,9_drop,9_drop,Hong and Kacperczyk,2009,Sin Stock (original list),JFE,,Other,other,1926,2006,sinOrig,9_drop,9_drop,,,,,,,,,,,"redundant,",0,3655.0
AbnormalAccrualsPercent,Placebo,indirect,,"Hafzalla, Lundholm, Van Winkle",2011,Percent Abnormal Accruals,AR,continuous,Accounting,accruals,1989,2008,AbnAccrPct,HXZ variant,,,,,,,,,,,,"OP says ""Cheng and Thomas (2005) assess 22 different abnomral accruals models … …However, all of their accruals models are scaled by some measure of firm size … never by earnings.""  They don't mention abnormal accruals anywhere else.",See AbnormalAccruals.  signal = AbnormalAccruals x lagged assets / absolute value of ni,211.0
AccrualQuality,Placebo,indirect,,"Francis, LaFond, Olsson, Schipper",2005,Accrual Quality,JAE,continuous,Accounting,accruals,1971,2002,AccrualQuality,correlated with E/P and factor structure,2 and 3,,,,,,,,,,,"Table 2 studies correlation between cost of debt proxies (E/P, beta) and accrual quality.  Table 3 regresses stock returns on contemporaneous accrual LS port returns.","Define Accruals (tempAccruals) as (difference between current assets (act) and one-year lagged current assets) - (difference between cash and short-term investments (che) and one-year lagged cash and short-term investments) - (difference between current liabilities (lct) and one-year lagged current liabilities) + (difference between debt in current liabilities (dlc) and one-year lagged debt in current liabilities) - (depreciation and amortization (dp)) all divided by (mean of total assets (at) and one-year lagged total assets ). Create tempCAcc as tempAccruals + dp/( (at + l.at)/2), tempRev as sale/( (at + l.at)/2), tempDelRev as tempRev - l.tempRev, tempPPE as ppegt/( (at + l.at)/2) where ppegt is total gross property, plant and equipment and tempCFO as ib/( (at + l.at)/2) - tempAccruals where ib is the income before extraordinary items. Run a regression for each year and industry of tempCAcc on the current value and one year lead and lag of tempDelRev and tempPPE. Save the regression residuals and replace with missing if there are not at least 20 obersvations per year and industry. Calculate accrual quality (AQ) as the standard deviation of residuals over 4 years. If more than one observation is missing set  AQ to missing. Replace AccrualQuality by the one-year lagged AQ to make sure the signal is available at time of investment.",4973.0
AccrualQualityJune,Placebo,indirect,,"Francis, LaFond, Olsson, Schipper",2005,Accrual Quality in June,JAE,continuous,Accounting,accruals,1971,2002,AccrualQualityJune,HXZ variant,,,,,,,,,,,,,see AccrualQuality. Update only with June values for each variable.,4973.0
AMq,Placebo,indirect,,Fama and French,1992,Total assets to market (quarterly),JF,continuous,Accounting,valuation,1975,1990,AMq,HXZ variant,,,,,,,,,,,,,Total assets (at) divided by market value of equity.,29636.0
AssetGrowth_q,Placebo,indirect,,"Cooper, Gulen and Schill",2008,Asset growth quarterly,JF,continuous,Accounting,investment,1968,2003,InvestAGq,HXZ variant,,,,,,,,,,,,,Annual growth rate of total assets (at),2228.0
AssetLiquidityBook,Placebo,indirect,,Ortiz-Molina and Phillips,2014,Asset liquidity over book assets,JFQA,continuous,Accounting,asset composition,1984,2006,AssetLiquidityBook,no predictability. Correlated with ICC,3 or 6,corr with ICC,,,,,,,,,,"Called WAIL3 in paper.  Not shown to predict returns; Tab 3 for main results, Tab 6 for ICC of WAIL,",(assets (at) + 0.75*(at - short term investments(che)) + 0.5*(at- curren total assets (act) - goodwill(gdwl) - intangibles (itan)) scaled by the one-month lagged book assets (at).,259.0
AssetLiquidityBookQuart,Placebo,indirect,,Ortiz-Molina and Phillips,2014,Asset liquidity over book (qtrly),JFQA,continuous,Accounting,asset composition,1984,2006,AssetLiquidityBookQuart,HXZ variant,,,,,,,,,,,,"Not shown to predict returns; Tab 3 for main results, Tab 6 for ICC of WAIL,",(short term investments (cheq) + 0.75*(actq - short term investments(cheq)) + 0.5*(atq- curren total assets (actq) - goodwill(gdwlq) - intangibles (itanq)) scaled by the one-month lagged book assets (atq). Replace goodwill and intangibles with 0 if they are missing.,259.0
AssetLiquidityMarket,Placebo,indirect,,Ortiz-Molina and Phillips,2014,Asset liquidity over market,JFQA,continuous,Accounting,asset composition,1984,2006,AssetLiquidityMarket,no predictability. Correlated with ICC,3 or 6,corr with ICC,,,,,,,,,,"Called MWAIL in paper.  Not shown to predict returns; Tab 3 for main results, Tab 6 for ICC of WAIL,",(assets (at) + 0.75*(at - short term investments(che)) + 0.5*(at- curren total assets (act) - goodwill(gdwl) - intangibles (itan)) scaled by the one-month lagged market assets (at + end-of-fiscal-year-stock-price (prcc_f) * common shares outstanding (csho) - book equity (ceq)).,259.0
AssetLiquidityMarketQuart,Placebo,indirect,,Ortiz-Molina and Phillips,2014,Asset liquidity over market (qtrly),JFQA,continuous,Accounting,asset composition,1984,2006,AssetLiquidityMarketQuart,HXZ variant,,,,,,,,,,,,"Not shown to predict returns; Tab 3 for main results, Tab 6 for ICC of WAIL,",(short term investments (cheq) + 0.75*(actq - short term investments(cheq)) + 0.5*(atq- curren total assets (actq) - goodwill(gdwlq) - intangibles (itanq)) scaled by the one-month lagged book assets (atq) + lagged closing price (prccq) times lagged common shares outstanding (cshoq) - lagged book equity (ceqq). Replace goodwill and intangibles with 0 if they are missing.,259.0
AssetTurnover,Placebo,indirect,1_good,Soliman,2008,Asset Turnover,AR,continuous,Accounting,composite accounting,1984,2002,ATurn,t=0.3 in mv reg,7 ATO,mv reg,1.0,,0.29,EW,,,12.0,6.0,abs(prc)>5,No predictive power in multivariate predictive regression (table 7). Not clear how it would play out as univariate hedge strategy.,"Sales (sale) divided by two year average of net operating assets. Net operating assets is the sum of receivables (rect), inventories (invt), current assets other (aco), net property, plants and equipment (ppent) and intangibles (intan), minus accounts payable (ap), other current liabilities (lco) and other liabilities (lo). Exclude if abs(prc) < 5 or AssetTurnover < 0.",919.0
AssetTurnover_q,Placebo,indirect,,Soliman,2008,Asset Turnover,AR,continuous,Accounting,composite accounting,1984,2002,AssetTurnover_q,HXZ variant,,,,,,,,,,,,,"Sales (sale) divided by two year average of net operating assets. Net operating assets is the sum of receivables (rect), inventories (invt), current assets other (aco), net property, plants and equipment (ppent) and intangibles (intan), minus accounts payable (ap), other current liabilities (lco) and other liabilities (lo). Exclude if abs(prc) < 5 or AssetTurnover < 0.",919.0
betaCC,Placebo,indirect,2_fair,Acharya and Pedersen,2005,Illiquidity-illiquidity beta (beta2i),JFE,continuous,Trading,liquidity,1964,1999,betaCC,in-sample only,"Tab 4, Fig 2",in-samp GMM,1.0,,,,,,,,,"Table 4 fits an expected return equation but using a ""GMM framework that takes into account the pre-estimation of betas,"" but this seems to be an in-sample fit only.",see monthly Code,4897.0
betaCR,Placebo,indirect,2_fair,Acharya and Pedersen,2005,Illiquidity-market return beta (beta4i),JFE,continuous,Trading,liquidity,1964,1999,betaCR,in-sample only,"Tab 4, Fig 2",in-samp GMM,1.0,,,,,,,,,"Table 4 fits an expected return equation but using a ""GMM framework that takes into account the pre-estimation of betas,"" but this seems to be an in-sample fit only.",see monthly Code,4897.0
BetaDimson,Placebo,indirect,,Dimson,1979,Dimson Beta,JFE,continuous,Price,market risk,1955,1974,BetaDimson,only shown to forecast beta,,,,,,,,,,,,"Not shown to predict returns.  Whole paper is just about forecasting beta,","Rolling regression of daily return (ret - rf) on the same-day, one-day ahead, and one-day lagged value of the market return (mktrf). Rolling regression with 20 observations (minimum 15). BetaDimson is the sum of the three coefficients.",3947.0
betaNet,Placebo,indirect,2_fair,Acharya and Pedersen,2005,"Net liquidity beta (betanet,p)",JFE,continuous,Trading,liquidity,1964,1999,betaNet,in-sample only,"Tab 4, Fig 2",in-samp GMM,1.0,,,,,,,,,"Table 4 fits an expected return equation but using a ""GMM framework that takes into account the pre-estimation of betas,"" but this seems to be an in-sample fit only.",see monthly Code,4897.0
betaRC,Placebo,indirect,2_fair,Acharya and Pedersen,2005,Return-market illiquidity beta,JFE,continuous,Trading,liquidity,1964,1999,betaRC,in-sample only,"Tab 4, Fig 2",in-samp GMM,1.0,,,,,,,,,"Table 4 fits an expected return equation but using a ""GMM framework that takes into account the pre-estimation of betas,"" but this seems to be an in-sample fit only.",see monthly Code,4897.0
betaRR,Placebo,indirect,2_fair,Acharya and Pedersen,2005,Return-market return illiquidity beta,JFE,continuous,Trading,liquidity,1964,1999,betaRR,in-sample only,"Tab 4, Fig 2",in-samp GMM,1.0,,,,,,,,,"Table 4 fits an expected return equation but using a ""GMM framework that takes into account the pre-estimation of betas,"" but this seems to be an in-sample fit only.",see monthly Code,4897.0
BMq,Placebo,indirect,,"Rosenberg, Reid, and Lanstein",1985,Book to market (quarterly),JPM,continuous,Accounting,valuation,1973,1984,BMq,HXZ variant,,,,,,,,,,,,,Log of quarterly book equity over market equity.,32.0
BookLeverageQuarterly,Placebo,indirect,,Fama and French,1992,Book leverage (quarterly),JF,continuous,Accounting,leverage,1973,1990,BookLevq,HXZ variant,,,,,,,,,,,,,"Total assets (atq) divided by book value of equity plus deferred taxes (txditcq) and preferred stock. Equity is shareholder equity (seqq) if available, or book equity (ceqq) plus preferred stock (pstkq), or total assets minus total liabilities (ltq).",29636.0
BrandCapital,Placebo,indirect,,"Belo, Lin and Vitorino",2014,Brand capital to assets,RED,continuous,Accounting,investment alt,1975,2010,BrandCapital,not studied for predictability,,,,,,,,,,,,Not studied in OP.  OP studies BrandInvest.,"Brand capital is computed by the perpetual inventory method. In the first year, brand capital is advertising expense divided by (.5 + .1). In subsequent years, we let brand capital depreciate with a rate of .5 and add current advertising expenses. Brand capital is scaled by total assets (at). Set to missing if advertising expense is missing.",285.0
CapTurnover,Placebo,indirect,2_fair,Haugen and Baker,1996,Capital turnover,JFE,continuous,Accounting,turnover,1979,1993,CapTurn,t<2 in mv reg nonstandard,1 footnote,mv reg nonstandard,1.0,,,EW,0.2,,1.0,6.0,exchcd==1,"weak in original paper.  They forecast returns using many variables.  Capital turnover is not in the top 11 predictors,",Lagged sales (sale) divided by two-year lagged assets (at).,1647.0
CapTurnover_q,Placebo,indirect,,Haugen and Baker,1996,Capital turnover (quarterly),JFE,continuous,Accounting,turnover,1979,1993,CapTurnq,HXZ variant,,,,,,,,,,,,,Sales (saleq) divided by one quarter lagged assets (atq).,1647.0
cashdebt,Placebo,indirect,2_fair,Ou and Penman,1989,CF to debt,JAR,continuous,Accounting,profitability alt,1973,1983,cashdebt,ingredient in complicated model,3 and 5,LS from complicated model,,,,,,,,6.0,,"probably weak.  Paper uses many variables to forecast earnings, model is then used to forecast returns,",Sum of income before extraordinary items (ib) and depreciation and amortization (dp) divided by total liabilities (lt) averaged over the current and previous fiscal years.,1840.0
CBOperProfLagAT,Placebo,indirect,,Ball et al.,2016,Cash-based oper prof lagged assets,JFE,continuous,Accounting,profitability,1963,2014,ProfCashLag,HXZ variant,,,,,,,,,,,,"This is operating prof with an R&D adjustment.  Strong in FM reg with controls, but insignificant in VW port sorts (Tab 4). OP states they lag denominator, but 2015 JFE does not lag, no clear motivation to lag, and our replications much closer to OP without lag.","Revenue (revt) minus cost (cogs) - (administrative expenses (xsga) - R&D expenses (xrd))  minus annual change in receivables (rect), annual change in investment (invt) and annual change in prepaid expenses, plus annual change in current deferred revenue (drc), long-term deferred revenue (drlt), accounts payable (ap) and accrued expenses (xacc), all divided by total assets (at) in year t-1. Replace all variables in the numerator with 0 if they are missing. Exclude if share code is greater 11, market value of equity, BM or total assets are missing, or if SIC code between 6000 and 6999.",534.0
CBOperProfLagAT_q,Placebo,indirect,,Ball et al.,2016,Cash-based oper prof lagged assets qtrly,JFE,continuous,Accounting,profitability,1963,2014,ProfCashLagq,HXZ variant,,,,,,,,,,,,,"Revenue (revtq) minus cost (cogsq) - (administrative expenses (xsgaq) - R&D expenses (xrdq))  minus annual change in receivables (rectq), annual change in investment (invtq) and annual change in prepaid expenses, plus annual change in current deferred revenue (drcq), long-term deferred revenue (drlt), accounts payable (apq) and accrued expenses (xaccq), all divided by total assets (atq) in previous quarter. Replace all variables in the numerator with 0 if they are missing. Exclude if share code is greater 11, market value of equity, BM or total assets are missing, or if SIC code between 6000 and 6999.",534.0
cfpq,Placebo,indirect,,"Desai, Rajgopal, Venkatachalam",2004,Operating Cash flows to price quarterly,AR,continuous,Accounting,valuation,1973,1997,CFOper2Priceq,HXZ variant,,,,,,,,,,,,,"Operating cash-flow (oancf) divided by market value of equity. If operating cash-flow is missing, replace by difference betwee net income (ib) and level of accruals, where the latter is the annual change in current assets (act) minus the annual change in cash and short-term investments (che), minus the annual change in current liabilities (lct) plus the annual change in debt in current liabilities (dlc) plus the annual change in payable income taxes (txp) plus depreciation (dp).",650.0
CFq,Placebo,indirect,,"Lakonishok, Shleifer, Vishny",1994,Cash flow to market quarterly,JF,continuous,Accounting,valuation,1968,1990,CFq,HXZ variant,,,,,,,,,,,,,Net income (ib) plus depreciation (dp) divided by market equity. Exclude NASDAQ stocks.,7654.0
ChangeRoA,Placebo,indirect,,"Balakrishnan, Bartov and Faurel",2010,Change in Return on assets,,continuous,Accounting,composite accounting,1976,2005,ChangeRoA,HXZ variant,,,,,,,,,,,,"HXZ do not cite anyone for this ""replication.""  OP (HXZ) uses RDQ to get very timely data, but we just lag everything by 3 months for simplicity.  This is pretty much a combination of profitability and investment, if you think about it.",Quarterly return on assets (rdq/atq) minus its value four quarters ago.,338.0
ChangeRoE,Placebo,indirect,,"Balakrishnan, Bartov and Faurel",2010,Change in Return on equity,,continuous,Accounting,composite accounting,1976,2005,ChangeRoE,HXZ variant,,,,,,,,,,,,"HXZ do not cite anyone for this ""replication.""  OP (HXZ) uses RDQ to get very timely data, but we just lag everything by 3 months for simplicity.  This is pretty much a combination of profitability and investment, if you think about it.",Quarterly return on equity (ceqq/atq) minus its value four quarters ago.,338.0
ChNCOA,Placebo,indirect,1_good,Soliman,2008,Change in Noncurrent Operating Assets,AR,continuous,Accounting,investment,1984,2002,LTAssetGr,No predictability.  Ingredient for predictor.,7 footnote,,,,,,,,,,,,Twelve-month change in noncurrent operating assets. Noncurrent operating assets is ( (at - act - ivao) - (lt - dlc - dltt) )/at.,919.0
ChNCOL,Placebo,indirect,1_good,Soliman,2008,Change in Noncurrent Operating Liab,AR,continuous,Accounting,investment alt,1984,2002,ChNCOL,No predictability.  Ingredient for predictor.,7 footnote,,,,,,,,,,,,Twelve-month change in noncurrent operating liabilites. Noncurrent operating liabilites is (lt - dlc- dltt).,919.0
ChPM,Placebo,indirect,1_good,Soliman,2008,Change in Profit Margin,AR,continuous,Accounting,sales growth,1984,2002,PMGrowth,t=0.3 in mv reg,7 Model 1 DeltaPM,mv reg,1.0,,0.33,EW,,,12.0,6.0,abs(prc)>5,"Tab 7 Delta PM, insignificant with many controls, no sorts.",Annual change in profit margin PM (profit margin defined below). Exclude if price less than 5.,919.0
currat,Placebo,indirect,2_fair,Ou and Penman,1989,Current Ratio,JAR,continuous,Accounting,asset composition,1973,1983,currat,ingredient in complicated model,3 and 5,LS from complicated model,,,,,,,,6.0,,"probably weak.  Paper uses many variables to forecast earnings, model is then used to forecast returns,","Current total assets (act) divided by current total liabilities (lct). If missing, replace current total assets by cash and short-term investments (che) plus total receivables (rect) plus total inventory (invt), and replace current total liabilities (lct) with accounts payable (ap).",614.0
depr,Placebo,indirect,2_fair,Holthausen and Larcker,1992,Depreciation to PPE,JAE,continuous,Accounting,other,1978,1988,depr,ingredient in complicated model,Table 3,port sort on massive model,,,,,,,,,,"Paper fits a massive model to returns (Table 2), then sorts stocks based on predictions (Table 3).  Depreciation is mentioned in Footnote 6, which says that depreciation enters into the models eight times.","Depreciation and amortization (dp) divided by property, plant and equipment net total (ppent).",466.0
DivYield,Placebo,indirect,1_good,"Naranjo, Nimalendran, Ryngaert",1998,Dividend yield for small stocks,JF,continuous,Accounting,valuation,1963,1994,DivYield,"mixed results, small spread",1B,port sort no LS,1.0,,,EW,0.25,,1.0,12.0,"abs(prc)>2, exchcd%in%c(1)","Ret is nonmonotic in DivYield, except in small stocks, as seen in Tab 1B.  FF3 adjusted returns are monotonic however due to size adjustment (Tab 3).  Previous papers (Keim 1985) also find mixed results, this paper is a more careful look.  Unclear if likely or maybe predictor.  Perhaps Keim (1985) or Bloom (1980) or even an earlier paper could be cited, but these earlier papers don't show convincing predictability in our view.  See also Kalay and Michaely (2000)",Define tempDY as 4 times latest dividend (divamt) divided by price (prc). Define positive yield stocks as those which paid a dividend in all of the past 4 quarters.  Set DivYield to missing if stock is above the median firm size.  This procedure is based on Table 1B.,265.0
DivYieldAnn,Placebo,indirect,,"Naranjo, Nimalendran, Ryngaert",1998,Last year's dividends over price,,continuous,Accounting,valuation,1963,1994,DivYieldAnn,HXZ variant,,,,,,,,,,,,See DivYieldST,,265.0
EarningsConservatism,Placebo,indirect,,"Francis, LaFond, Olsson, Schipper",2004,Earnings conservatism,AR,continuous,Accounting,other,1975,2001,EarningsConservatism,correlated with BM and other predictors,5,,,,,,,,,,,Table 5 regresses cost of equity proxies (e.g. beta) on various earnings attributes.,"Earnings (ib) scaled by market capitalization (shrout*abs(prc)) regressed on the 15 month stock return from (t=-11 to t=+3), an indicator for whether that 15- month return is negative and the interaction of these two variables. Rolling regression in annual data with 10 observations. EarningsConservatism is the sum of the coefficients on the 15-month stock return and the interaction term divided by the coefficient on the 15-month stock return.",4420.0
EarningsPersistence,Placebo,indirect,,"Francis, LaFond, Olsson, Schipper",2004,Earnings persistence,AR,continuous,Accounting,other,1975,2001,EarningsPersistence,correlated with BM and other predictors,5,,,,,,,,,,,Table 5 regresses cost of equity proxies (e.g. beta) on various earnings attributes.,Earnings per share (epspx/ajex) regressed on its value from the previous year. Rolling regression in annual data with 10 observations. EarningsPersistence is the coefficient on the lagged value.,4420.0
EarningsPredictability,Placebo,indirect,,"Francis, LaFond, Olsson, Schipper",2004,Earnings Predictability,AR,continuous,Accounting,other,1975,2001,EarningsPredictability,correlated with BM and other predictors,5,,,,,,,,,,,Table 5 regresses cost of equity proxies (e.g. beta) on various earnings attributes.,Earnings per share (epspx/ajex) regressed on its value from the previous year. Rolling regression in annual data with 10 observations. EarningsPredictability is the R2 of this regression.,4420.0
EarningsSmoothness,Placebo,indirect,,"Francis, LaFond, Olsson, Schipper",2004,Earnings Smoothness,AR,continuous,Accounting,other,1975,2001,EarningsSmoothness,correlated with BM and other predictors,5,,,,,,,,,,,Table 5 regresses cost of equity proxies (e.g. beta) on various earnings attributes.,10-year rolling standard deviation of (ib/at) divided by 10-year rolling standard deviation of cash-flow ((ib - ( (act - act_{t-1}) - (lct - lct_{t-1}) - (che - che_{t-1}) + (dlc - dlc_{t-1}) - dp))/at_{t-1}),4420.0
EarningsTimeliness,Placebo,indirect,,"Francis, LaFond, Olsson, Schipper",2004,Earnings timeliness,AR,continuous,Accounting,other,1975,2001,EarningsTimeliness,correlated with BM and other predictors,5,,,,,,,,,,,Table 5 regresses cost of equity proxies (e.g. beta) on various earnings attributes.,"Earnings (ib) scaled by market capitalization (shrout*abs(prc)) regressed on the 15 month stock return from (t=-11 to t=+3), an indicator for whether that 15- month return is negative and the interaction of these two variables. Rolling regression in annual data with 10 observations. EarningsTimeliness is the R2 of this regression.",4420.0
EarningsValueRelevance,Placebo,indirect,,"Francis, LaFond, Olsson, Schipper",2004,Value relevance of earnings,AR,continuous,Accounting,other,1975,2001,EarningsValueRelevance,correlated with BM and other predictors,5,,,,,,,,,,,Table 5 regresses cost of equity proxies (e.g. beta) on various earnings attributes.,"15 month stock return from (t=-11 to t=+3) regressed on Earnings (ib) and the change in earnings, both scaled by market capitalization (shrout*abs(prc)). Rolling regression in annual data with 10 observations. EarningsValueRelevance is the R2 of this regression.",4420.0
EBM_q,Placebo,indirect,,"Penman, Richardson and Tuna",2007,Enterprise component of BM,JAR,continuous,Accounting,valuation,1963,2001,EBM_q,HXZ variant,,,,,,,,,,,,"need to adjust sample dates: even though paper says it begins in 1962, there is only 1 stock for the first 5 months of 1962,",(ceq + che - dltt - dlc - dc - dvpa+ tstkp ) / (mve\_c + che - dltt - dlc - dc - dvpa+ tstkp). Exclude if price less than 5.,551.0
EntMult_q,Placebo,indirect,,Loughran and Wellman,2011,Enterprise Multiple quarterly,JFQA,continuous,Accounting,valuation,1963,2009,EntMult_q,HXZ variant,,,,,,,,,,,,,"Market value of equity + long-term debt (dltt) + debt in current liabilities (dlc) + deferred charges (dc) - cash and short-term investments (che) , divided by operating income (oibdp). Exclude if missing book equity or negative operating income.",224.0
EPq,Placebo,indirect,,Basu,1977,Earnings-to-Price Ratio,JF,continuous,Accounting,valuation,1963,1971,EPq,HXZ variant,,,,,,,,,,,,,"ib / lag(market value of equity, 6 months).  NYSE stocks only.  Exclude if EP < 0.  Lag simulates the Dec 31 market equity used in original paper",5363.0
ETR,Placebo,indirect,1_good,Abarbanell and Bushee,1998,Effective Tax Rate,AR,continuous,Accounting,other,1974,1988,TaxRate,t=1.5 in mv reg,2b RETR,reg,-1.0,,1.468,EW,,,12.0,6.0,,,"Let Tax be tax payments (txt) over pre-tax income (pi) and amortization of intangibles (am, 0 if missing). Let earnings per share be epspx/ajex. ETR if Tax minus the average of Tax over the previous three years multiplied by one-year EPS growth scaled by by price (prcc_f).",1124.0
FailureProbabilityJune,Placebo,indirect,,"Campbell, Hilscher and Szilagyi",2008,Failure probability,JF,continuous,Accounting,default risk,1981,2003,FailurePrJune,HXZ variant,,,,,,,,,,,,Tab6 Mean LS 2080 t-stat 1.41. We previously used monthly holding periods and got large t-stats.,"Failure probability is -9.16 -.058*PRICE + .075*MB - 2.13*CASHMTA -.045*RSIZE + 1.41*IdioRisk - 7.13*EXRETAVG + 1.42*TLMTA - 20.26*NIMTAAVG. PRICE is log(min(abs(prc), 15)); MB is shrout*abs(prc)/ceqq; CASHMTA is cheq/(shrout*abs(prc) + ltq); RSIZE is log(shrout*abs(prc)/ sum of shrout*abs(prc) for the largest 500 companies each month); IdioRisk is defined above, EXRETAVG is the weighted average excess return (log(1 + ret) - log(1 + mktrf)) over the previous 12 months, with weight on month t-j being $\phi^j$ and the sum scaled by $\frac{1-\phi}{1-\phi^{12}}$; TLMTA is total liabilities (ltq/(shrout*abs(prc)); NIMTAAVG is a weighted average of net income over total assets (ibq/(shrout*abs(prc) + ltq)) over four quarters, with weight $\phi^q$ on quarter $t-q$ and the sum scaled by $\frac{1-\phi^3}{1-\phi^{12}}$. $\phi = 2^{-\frac{1}{3}}$. All input variables are winsorized at the 5th and 95th percentile. Exclude if price less than 1.",3359.0
fgr5yrNoLag,Placebo,indirect,,La Porta,1996,Long-term EPS forecast (Monthly),JF,continuous,Analyst,earnings forecast,1983,1990,EPSForeLT,HXZ variant,,,,,,,,,,,,"This is the simple and perhaps intuitive long term forecast signal, but OP is very specific about timing.","Long-term earnings forecast (fgr5yr).  Exclude if book equity (ceq), net income (ib), deferred taxes (txdi), dividends (dvp), revenue (sale) or depreciation (dp) is missing.",1469.0
FRbook,Placebo,indirect,1_good,Franzoni and Marin,2006,Pension Funding Status,JF,continuous,Accounting,composite accounting,1980,2002,PensionFundBook,HXZ variant,,,,,,,,,,,,Not studied in OP.,"FR = (FVPA - PBO), scaled by total assets (at). FVPA is pbnaa from 1980 to 1986, pplao + pplao from 1987 to 1997, and pplao after 1997. PBO is pbnvv from 1980 to 1986, pbpro + pbpru from 1987 to 1997, and pbpro after 1997. Exclude if price less than 5 or  shrcd > 11.",324.0
GPlag,Placebo,indirect,,Novy-Marx,2013,gross profits / total assets,JFE,continuous,Accounting,profitability,1963,2010,GPlag,HXZ variant,,,,,,,,,,,,HXZ variant,"Revenue (sale) - cost of goods solds (cogs), divided by 12 months lagged total assets.",2983.0
GPlag_q,Placebo,indirect,,Novy-Marx,2013,gross profits / total assets,JFE,continuous,Accounting,profitability,1963,2010,GPlag_q,HXZ variant,,,,,,,,,,,,HXZ variant,"Revenue (revtq) - cost of goods solds (cogsq), divided by one quarter lagged total assets (atq).",2983.0
grcapx1y,Placebo,indirect,,Anderson and Garcia-Feijoo,2006,Investment growth (1 year),AR,continuous,Accounting,investment,1964,2003,CAPXgr1y,HXZ variant,,,,,,,,,,,,"HXZ deviates from OP.  OP only examines (1) capx / lag(capx,2) and capx / sum( capx(t-3:t) ).  OP notation is odd, uses cegth2 and cegth3 but no cegth.","Growth between one-year lagged capital expenditures (capx) and two-year lagged capital expenditures. Replace capx with the one year difference in property, plant and equipment (ppent) if capx is missing and the corresponding firm age is greater or equal than two years.",588.0
GrGMToGrSales,Placebo,indirect,1_good,Abarbanell and Bushee,1998,Gross margin growth to sales growth,AR,continuous,Accounting,earnings growth,1974,1988,GM2SaleGr,t=1.9 in mv reg,2b RGM,reg,1.0,,1.857,EW,,,12.0,6.0,,,"Define gross margin GM as revenue (sale) minus cost of goods sold (cogs). GrGMToGrSales is the percentage growth of GM relative to average GM in years t-1 and t-2, minus the percentage growth of revenue relative to average revenue in years t-1 and t-2. Replace growth rates with growth relative to the previous year only if data for t-2 are not available.",1124.0
GrSaleToGrReceivables,Placebo,indirect,1_good,Abarbanell and Bushee,1998,Change in sales vs change in receiv,AR,continuous,Accounting,profitability alt,1974,1988,RevG2ARG,t=1.6 in mv reg,2b RAR,reg,1.0,,1.622,EW,,,12.0,6.0,,,"Percentage growth in sales (sale) relative to average sales of t-1 and t-2, minus percentage growth in receivables (rect) relative to average receivables of t-1 and t-2. Both growth terms are calculated relative to t-1 only if t-2 is missing.",11.0
IdioVolCAPM,Placebo,indirect,,Ang et al.,2006,Idiosyncratic risk (CAPM),JF,continuous,Price,volatility,1963,2000,IdioVolCAPM,HXZ variant,,,,,,,,,,,,,Standard deviation of residuals from CAPM regressions using the past month of daily data.  Value weighted,2321.0
IdioVolQF,Placebo,indirect,,Ang et al.,2006,Idiosyncratic risk (q factor),JF,continuous,Price,volatility,1967,2000,IdioVolQF,HXZ variant,,,,,,,,,,,,,Standard deviation of residuals from q-factor regressions using the past month of daily data.  Value weighted,2321.0
IntrinsicValue,Placebo,indirect,,Frankel and Lee,1998,Intrinsic or historical value,JAE,continuous,Accounting,valuation,1975,1993,IntrinsicValue,not studied.  Ingredient variable.,,,,,,,,,,,,"Called V_h in paper or historical earnings based value in paper.  Our name comes from HXZ, but we should probably rename it.","Value based on a two-stage dividend discount model assuming ROE is remains the same as the most recent observation, scaled by market value.",1741.0
KZ_q,Placebo,indirect,,"Lamont, Polk and Saa-Requejo",2001,Kaplan Zingales index quarterly,RFS,continuous,Accounting,composite accounting,1968,1997,KZ_q,HXZ variant,,,,,,,,,,,,,-1.002* (net income (ni) + depreciation (dp))/total assets (at) + .283*(total assets (at) + market value of equity - book value of equity (ceq) - deferred taxes (txdi))/total assets (at) + 3.319*(debt in current liabilities (dlc) + long-term debt (dltt))/(debt in current liabilities + long-term debt + book value of equity) - 39.368*(Dividends (divamt)/total assets) - 1.315*(cash and short-term investments (che)/total assets). Replace txdi and divamt with 0 if missing.,24.0
LaborforceEfficiency,Placebo,indirect,1_good,Abarbanell and Bushee,1998,Laborforce efficiency,AR,continuous,Accounting,profitability alt,1974,1988,RevG2EmpG,t=0.6 in mv reg,2b RLF,reg,1.0,,0.61,EW,,,12.0,6.0,,,One year growth of the ratio between sales (sale) and employment (emp).,1124.0
Leverage_q,Placebo,indirect,,Bhandari,1988,Market leverage quarterly,JFE,continuous,Accounting,leverage,1952,1981,Leverageq,HXZ variant,,,,,,,,,,,,,Total liabilities (lt) divided by market value of equity.,2596.0
nanalyst,Placebo,indirect,1_good,"Elgers, Lo and Pfeiffer",2001,Number of analysts,AR,continuous,Analyst,info proxy,1982,1998,nanalyst,spread in median ret each leg size adj,2 SAR_t,summary stat,1.0,0.566666667,,EW,0.5,,1.0,12.0,,"Not a focus of the paper.  Table 2 shows summary stats that suggest size adjusted returns are different by group, but no t-stats and size adjustment may affect each group differently.",Number of estimates (numest) in IBES for one-quarter ahead earnings. Replace with 0 if missing after 1989.,293.0
NetDebtPrice_q,Placebo,indirect,,"Penman, Richardson and Tuna",2007,Net debt to price,JAR,continuous,Accounting,leverage,1963,2001,NetDebtPrice_q,HXZ variant,,,,,,,,,,,,"need to adjust sample dates: even though paper says it begins in 1962, there is only 1 stock for the first 5 months of 1962,","Long-term debt (dltt) plus debt in current liabilities (dlc) plus preferred stock (pstk) plus preferred dividends in arrears (dvpa) minus treasury stock (tstkp) minus cash and short-term investments (che), scaled by market value of equity. Exclude if SIC between 6000 and 6999, or if missing value for total assets (at), net income (ib), common shares outstanding (csho), book value of equity (ceq) or price close fiscal year (prcc\_f).  Keep only 3rd B/M Quintile, following Table 4 (and in contrast to Table 1).",551.0
NetPayoutYield_q,Placebo,indirect,,Boudoukh et al.,2007,Net Payout Yield quarterly,JF,continuous,Accounting,valuation,1984,2003,NPayYieldq,HXZ variant,,,,,,,,,,,,,"Quarterly dividends tempDiv = dvpsxq*cshoq*ajexq (zero if dvpsxq is exactly zero; otherwise require all inputs). Gross payout is tempDiv plus individual-quarter purchases (prstkcyq) plus max(pstkq lagged 3 months minus pstkq, 0). Equity issuance is individual-quarter sales (sstkyq) minus max(pstkq minus pstkq lagged 3 months, 0). NetPayoutYield_q is gross payout minus equity issuance, divided by current mve_permco. Require finite nonzero yield and finite positive current market cap. An unavailable preferred-stock difference contributes zero to each adjustment. Use matched accounting rows; preserve missing flows without extra imputation.",777.0
OperProfLag,Placebo,indirect,,Fama and French,2006,operating profits / book equity,JFE,continuous,Accounting,profitability,1977,2003,ProfOperLag,HXZ variant,,,,,,,,,,,,,"Revenue (revt) minus cost (cogs) - administrative expenses (xsga) - interest expenses (xint), scaled by lagged book value of equity (ceq). Exclude smallest size tercile.",1599.0
OperProfLag_q,Placebo,indirect,,Fama and French,2006,operating profits / book equity,JFE,continuous,Accounting,profitability,1977,2003,ProfOperLagq,HXZ variant,,,,,,,,,,,,,"Revenue (revtq) minus cost (cogsq) - administrative expenses (xsgaq) - interest expenses (xintq), scaled by equity. Equity is shareholder equity (seqq) if not missing. Otherwise, the sum of book equity (ceqq) and preferred stock (pstkq) if not missing. Otherwise, assets (atq) minus liabilities (ltq).",1599.0
OperProfRDLagAT,Placebo,indirect,,Ball et al.,2016,Oper prof R&D adj lagged assets,JFE,continuous,Accounting,profitability,1963,2014,OperProfRDLag,HXZ variant,,,,,,,,,,,,,"Revenue (revt) minus cost (cogs) - (administrative expenses (xsga) - R&D expenses (xrd)), all divided by total assets (at) in year t-1. Replace all variables in the numerator with 0 if they are missing. Exclude if share code is greater 11, market value of equity, BM or total assets are missing, or if SIC code between 6000 and 6999.",534.0
OperProfRDLagAT_q,Placebo,indirect,,Ball et al.,2016,Oper prof R&D adj lagged assets (qtrly),JFE,continuous,Accounting,profitability,1963,2014,OperProfRDLagAT_q,HXZ variant,,,,,,,,,,,,,"Revenue (revtq) minus cost (cogsq) - (administrative expenses (xsgaq) - R&D expenses (xrdq)), all divided by total assets (atq) in quarter t-1. Replace all variables in the numerator with 0 if they are missing. Exclude if share code is greater 11, market value of equity, BM or total assets are missing, or if SIC code between 6000 and 6999.",534.0
OPLeverage_q,Placebo,indirect,,Novy-Marx,2011,Operating leverage (qtrly),ROF,continuous,Accounting,other,1963,2008,OPLeverage_q,HXZ variant,,,,,,,,,,,,,"Sum of administrative expenses (xsga) and cost of goods sold (cogs), scaled by total assets (at).  Use xsga = 0 if xsga is missing.",533.0
OrgCapNoAdj,Placebo,indirect,,Eisfeldt and Papanikolaou,2013,Org cap w/o industry adjustment,JF,continuous,Accounting,R&D,1970,2008,OrgCapNoAdj,HXZ variant,,,,,,,,,,,,see OrgCapAdj,OrgCapAdj without the industry adjustment (see OrgCapAdj).,1211.0
OScore_q,Placebo,indirect,,Dichev,1998,O Score quarterly,JF,discrete,Accounting,default risk,1981,1995,OScoreq,HXZ variant,,,,,,,,,,,,,"OScore = -1.32 - .407*log(at/GNP deflator) + 6.03*(lt/at) - 1.43*( (act - lct)/at) + .076*(lct/act) - 1.72*I(lt > at) - 2.37*(ib/at) - 1.83*(fopt/lt) + .285*(ib + ib$_{t-12}$ + ib$_{t-24}$ < 0) -  .521*( (ib - ib$_{t-12}$)/(abs(ib) + .abs(ib$_{t-12}$)) ). fopt = oancf if fopt is missing.  Exclude Exclude if SIC code between 3999 and 4999, or greater than 5999. Exclude if price less than 5.  Then exclude if OScore is in bottom quintile of OScore (original paper shows non-monotonic returns, as does our replication)",1614.0
PayoutYield_q,Placebo,indirect,,Boudoukh et al.,2007,Payout Yield quarterly,JF,continuous,Accounting,valuation,1984,2003,PayYieldq,HXZ variant,,,,,,,,,,,,,"Quarterly dividends tempDiv = dvpsxq*cshoq*ajexq (zero if dvpsxq is exactly zero; otherwise require all inputs). Gross payout is tempDiv plus individual-quarter purchases (prstkcyq) plus max(pstkq lagged 3 months minus pstkq, 0). PayoutYield_q is gross payout divided by mve_permco lagged 6 months. Require finite positive yield and finite positive lagged market cap. An unavailable preferred-stock difference contributes zero to each adjustment. Use matched accounting rows; preserve missing flows without extra imputation.",777.0
pchcurrat,Placebo,indirect,2_fair,Ou and Penman,1989,Change in Current Ratio,JAR,continuous,Accounting,investment alt,1973,1983,pchcurrat,ingredient in complicated model,3 and 5,LS from complicated model,,,,,,,,6.0,,"probably weak.  Paper uses many variables to forecast earnings, model is then used to forecast returns,",One year growth of the current ratio (currat) defined as current total assets (act) divided by current total liabilities (lct).,616.0
pchdepr,Placebo,indirect,2_fair,Holthausen and Larcker,1992,Change in depreciation to PPE,JAE,continuous,Accounting,investment alt,1978,1988,DepGr,ingredient in complicated model,Table 3,port sort on massive model,,,,,,,,,,"Paper fits a massive model to returns (Table 2), then sorts stocks based on predictions (Table 3).  Footnote 6 says depreciation divided by plant assets enters the models positively six times and negatively twice.","Annual percentage change in the ratio of depreciation (dp) to property, plant and equipment (ppent).",466.0
pchgm_pchsale,Placebo,indirect,,Abarbanell and Bushee,1998,Change in gross margin vs sales,AR,continuous,Accounting,profitability alt,1974,1988,ChAssetTurnover,GHZ variant of GrGMToGrSale,,,,,,,,,,,,GrGMtoGrSales is closer to OP,growth in (sale-cogs) minus growth in sale,1124.0
pchquick,Placebo,indirect,2_fair,Ou and Penman,1989,Change in quick ratio,JAR,continuous,Accounting,investment alt,1973,1983,pchquick,ingredient in complicated model,3 and 5,LS from complicated model,,,,,,,,6.0,,"probably weak.  Paper uses many variables to forecast earnings, model is then used to forecast returns,","One year growth in quick ratio, defined as change between total current assets (act) and inventory (invt) scaled by the total current liabilities (lct).",616.0
pchsaleinv,Placebo,indirect,2_fair,Ou and Penman,1989,Change in sales to inventory,JAR,continuous,Accounting,profitability alt,1973,1983,pchsaleinv,ingredient in complicated model,3 and 5,LS from complicated model,,,,,,,,6.0,,"probably weak.  Paper uses many variables to forecast earnings, model is then used to forecast returns,",One year growth in sales (sale) to inventory (invt) ratio.,616.0
PM,Placebo,indirect,1_good,Soliman,2008,Profit Margin,AR,continuous,Accounting,profitability,1984,2002,ProfitMargin,t=1 in mv reg,7:00 PM,mv reg,1.0,,1.04,EW,,,12.0,6.0,abs(prc)>5,No predictive power in multivariate predictive regression (table 7). Not clear how it would play out as univariate hedge strategy.,Net income (ni) over revenue (revt). Exclude if price less than 5.,920.0
PM_q,Placebo,indirect,,Soliman,2008,Profit Margin,AR,continuous,Accounting,profitability,1984,2002,PM_q,HXZ variant,,,,,,,,,,,,,Net income (ni) over revenue (revt). Exclude if price less than 5.,920.0
PS_q,Placebo,indirect,,Piotroski,2000,Piotroski F-score,JAR,continuous,Accounting,composite accounting,1976,1996,PS_q,HXZ variant,,,,,,,,,,,,"works very well if you value-weight! OP does not explicitly say if it is equal or value weighting, but it has extremely strong results similar to our VW,",Sum of nine indicator variables which are: 1 if net income (ib) greater 0; 1 if net cash flow (oancf) greater 0; 1 if return on assets (ib/at) increased relative to previous year; 1 if net cash flow greater net income; 1 if long-term debt to assets (dltt/at) declined over the previous year;  if current assets to current liabilities (act/lct) increased over the previous year; 1 if ebit/sale (ebit = ib + txt + xint) increased over the previous year; 1 if revenue to assets increased over the previous year; 1 if shrout $\leq$ shrout last year. Include highest quintile of book-to-market only.  Exclude if missing any of the input variables.,1956.0
quick,Placebo,indirect,2_fair,Ou and Penman,1989,Quick ratio,JAR,continuous,Accounting,asset composition,1973,1983,quick,ingredient in complicated model,3 and 5,LS from complicated model,,,,,,,,6.0,,"probably weak.  Paper uses many variables to forecast earnings, model is then used to forecast returns,",Difference between current total assets (act) and total inventory (invt) all divided by the current total liabilities (lct).,616.0
RD_q,Placebo,indirect,,"Chan, Lakonishok and Sougiannis",2001,R&D over market cap quarterly,JF,continuous,Accounting,R&D,1975,1995,RDq,HXZ variant,,,,,,,,,,,,,R&D expense (xrd) over market value of equity.,2982.0
rd_sale_q,Placebo,indirect,,"Chan, Lakonishok and Sougiannis",2001,R&D to sales,JF,continuous,Accounting,R&D,1975,1995,rd_sale_q,HXZ variant,,,,,,,,,,,,,One year lagged quarterly R&D (xrdq) divided by one year lagged quarterly sales (saleq).,2982.0
ResidualMomentum6m,Placebo,indirect,,"Blitz, Huij and Martens",2011,6 month residual momentum,JEmpFin,continuous,Price,momentum,1930,2009,MomRes6m,HXZ variant,,,,,,,,,,,,"OP only uses 12m version.  It turns out that 6m residual momentum is very much like the standard 12m version, but I suppose it could have turned out differently.","Run a rolling regression over 36 months of excess return (retrf) on excess market return (mktrf), size and value factors (smb, hml) and compute idiosyncratic returns as the one-month lagged residual. ResidualMomentum is the rolling mean of the residual divided by the rolling standard deviation of the residual, both computed over the past 6 months.",2982.0
RetNOA,Placebo,indirect,1_good,Soliman,2008,Return on Net Operating Assets,AR,continuous,Accounting,composite accounting,1984,2002,RetNOA,t=1.4 in mv reg,7 RNOA,mv reg,1.0,,1.41,EW,,,1.0,6.0,abs(prc)>5,"Tab 7 RNOA, insignificant with many controls,",Net income (ni) over revenue (revt). Exclude if price less than 5.,908.0
RetNOA_q,Placebo,indirect,,Soliman,2008,Return on Net Operating Assets,AR,continuous,Accounting,composite accounting,1984,2002,RetNOA_q,HXZ variant,,,,,,,,,,,,,Net income (ni) over revenue (revt). Exclude if price less than 5.,908.0
ReturnSkewCAPM,Placebo,indirect,,"Bali, Engle and Murray",2015,Idiosyncratic skewness (CAPM),Book,continuous,Price,risk,1963,2012,RetskewCAPM,HXZ variant,,,,,,,,,,,,OP uses 3F,Skewness of idiosyncratic returns computed as residuals from regression of daily excess returns (ret - rf) on market factor (mktrf) over the previous month. We require at least 15 non-missing observations.,2.0
ReturnSkewQF,Placebo,indirect,,"Bali, Engle and Murray",2015,Idiosyncratic skewness (Q model),Book,continuous,Price,risk,1967,2012,RetSkewQF,HXZ variant,,,,,,,,,,,,OP uses 3F,"Skewness of idiosyncratic returns computed as residuals from regression of daily excess returns (ret - rf) on q-factors (r\_mkt, r\_me, r\_ia, r\_roe) over the previous month. We require at least 15 non-missing observations. We download q-factor data from \url{http://global-q.org/index.html}.",2.0
roavol,Placebo,indirect,,"Francis, LaFond, Olsson, Schipper",2004,RoA volatility,AR,continuous,Accounting,cash flow risk,1975,2001,roavol,correlated with BM and other predictors,5,,,,,,,,,,,Table 5 regresses cost of equity proxies (e.g. beta) on various earnings attributes.,Rolling standard deviation of quarterly return on assets (roaq) over 4 years (minimum 2 years).,4424.0
salecash,Placebo,indirect,2_fair,Ou and Penman,1989,Sales to cash ratio,JAR,continuous,Accounting,profitability alt,1973,1983,salecash,ingredient in complicated model,3 and 5,LS from complicated model,,,,,,,,6.0,,"probably weak.  Paper uses many variables to forecast earnings, model is then used to forecast returns,",Sales (sale) divided by cash and short-term investments (che).,616.0
saleinv,Placebo,indirect,2_fair,Ou and Penman,1989,Sales to inventory,JAR,continuous,Accounting,profitability alt,1973,1983,saleinv,ingredient in complicated model,3 and 5,LS from complicated model,,,,,,,,6.0,,"probably weak.  Paper uses many variables to forecast earnings, model is then used to forecast returns,",Sales (sale) divded by total inventory (invt).,616.0
salerec,Placebo,indirect,2_fair,Ou and Penman,1989,Sales to receivables,JAR,continuous,Accounting,profitability alt,1973,1983,salerec,ingredient in complicated model,3 and 5,LS from complicated model,,,,,,,,6.0,,"probably weak.  Paper uses many variables to forecast earnings, model is then used to forecast returns,",Sales (sale) divded by total receivables (rect).,616.0
secured,Placebo,indirect,1_good,Valta,2016,Secured debt,JFQA,continuous,Accounting,external financing,1985,2012,secured,t > 1.96 in mv reg,4A secured,mv reg,1.0,,1.96,EW,,,1.0,6.0,,Tab 5B shows sign of predictability depends on Z-score quartile.  One could improve this portfolio by focusing on low Z-score.,Debt/mortgages and other secured (dm) divided by long-term liabilities (dltt) plus current liabilities (dlc). Replace with 0 if missing.,125.0
securedind,Placebo,indirect,,Valta,2016,Secured debt indicator,JFQA,discrete,Accounting,external financing,1985,2012,securedind,GHZ variant,,,,,,,,,,,,Doesn't seem to be in the paper.,"Binary version of secured. 1 if secured greater than 0, 0 otherwise.",125.0
sgr,Placebo,indirect,,"Lakonishok, Shleifer, Vishny",1994,Annual sales growth,JF,continuous,Accounting,sales growth,1968,1990,sgr,HXZ variant,,,,,,,,,,,,"Variation of meanrankrevgrowth.  Original paper uses only 5-year growth.  Included to match HXZ,",Sales (sale) relative to t-1.,7670.0
sgr_q,Placebo,indirect,,"Lakonishok, Shleifer, Vishny",1994,Annual sales growth quarterly,JF,continuous,Accounting,sales growth,1968,1990,sgr_q,HXZ variant,,,,,,,,,,,,,Sales (sale) relative to t-1.,7670.0
SP_q,Placebo,indirect,,"Barbee, Mukherji and Raines",1996,Sales-to-price quarterly,FAJ,continuous,Accounting,valuation,1979,1991,Rev2Priceq,HXZ variant,,,,,,,,,,,,,Ratio of annual sales (sale) to market value of equity.,421.0
tang_q,Placebo,indirect,,Hahn and Lee,2009,Tangibility quarterly,JF,continuous,Accounting,asset composition,1973,2001,Tangibilityq,HXZ variant,,,,,,,,,,,,,"Cash and short-term investments (che) plus .715*receivables (rect) + .547*inventory (invt) + .535* property, plant and equipment (ppent), scaled by total assets (at). Only defined for manufacturing firms (SIC $\geq$ 2000 and SIC <4000). Exclude the lowest tercile of manufacturing firms by total assets.",244.0
Tax_q,Placebo,indirect,,Lev and Nissim,2004,Taxable income to income (qtrly),AR,continuous,Accounting,other,1973,2000,Tax_q,HXZ variant,,,,,,,,,,,,,"Ratio of Taxes paid and tax share of net income. Numerator is defined as the sum of foreign (txfo) and federal (txfed) income taxes. If either one is missing, numerator is defined as total taxes (txt) minus deferred taxes (txdi). Denominator is the product of the prevailing tax rate and net income (ib). Tax rate is .48 before 1979, .46 from 1979 to 1986, .4 in 1987, .34 between 1988 and 1992 and .35 from 1993 onwards. If net income is negative, and the numerator is positive, tax is defined as 1. Exclude if price less than 5.",873.0
WW_Q,Placebo,indirect,,Whited and Wu,2006,Whited-Wu index,RFS,continuous,Accounting,external financing,1975,2001,WW_Q,HXZ variant,,,,,,,,,,,,"Insignificant in original paper,",Group data by 3 digit SIC code and month to compute total sales (saleq) by industry. Calculate the Whited-Wu index as -0.091*(quarterly income before extraordninary items (ibq) + quarterly depreciation and amortizarion (dpq)) / (quarterly assets(atq)) - 0.062( if the quarterly dividends per share (dvpsxq) is not missing and greater than 0) + 0.021 * quarterly long-term debt (dlttq) / atq - 0.044 * log(atq) + 0.102 * (four-quarter growth of quarterly Industry Sales) - 0.035 * (one-quarter growth of quarterly sales (saleq)).,4320.0
ZScore_q,Placebo,indirect,,Dichev,1998,Altman Z-Score quarterly,JFE,continuous,Accounting,default risk,1981,1995,ZScoreq,HXZ variant,,,,,,,,,,,,,"1.2*(current assets (act) - current liabilities (lct))/total assets (at) + 1.4*(Retained earnings (re)/total assets (at)) + 3.3*(net income (ni) + interest expense (xint) + total taxes (txt))/total assets (at) +  .6*(market value of equity/Total liabilities (lt)) + revenue (revt)/ total assets (at). Include only NYSE stocks. Exclude if SIC code between 4000 and 4999, or above 5999.  Exclude if ZScore is in bottom quintile of ZScore (original paper shows non-monotonic returns, as does our replication)",1618.0