Hi Just exploring the dataset, for the daily portfolios, I just want to make sure that I am using the most up to date data -- this one only goes up and till early 2020 for example.

Signals/Code/DataDownloads/C_CompustatQuarterly.do

gen time_avail_m = mofd(datadate) + 3  // Assume data available with a 3 month lag

This may have two minor issues. If we compare the RDQ (which is the release date of earnings) and datadate:

1. RDQ < datadate: more than 600 obsrvations.
2. RDQ > datadate+90 (day): more than 50,000 observations (that are not public available with 90-day lag)

For above two scenarios, is it better to set epspxq and other quarterly accounting information to missing?

RSCRIPT_PATH does not need to be set up for linux systems provided that Rscript is set up

Hi,

I tried to run the Stata code but encountered some problems. I had these errors of variables not found: "_temp not found" for some variables. They are sd10_tempEarnings, mean6_temp, and sd60_tempEpsDiff. I got stuck here and failed to solve it. Could anyone please help?

If this issue is not actually an issue, please accept my apology. I am very new to Stata programming.

Thanks

Thanks for your project and very appreciate your work to make it open source.

Can I ask which package are you using to extract FRED data? Is this one as below?
https://github.com/sboysel/fredr

If so, there is an open issue for this package: sboysel/fredr#75

install.packages("fredr") is currently not working and should install this package from github: devtools::install_github("sboysel/fredr")

Enhance documentation by adding link about "ssh to WRDS" in /Signals/Code/PrepScripts/masterSAS.sas.

ShareVol.do actually drops too many observations because it does not properly deal with observations at the beginning of the sample (where there aren't enough lags of the variables).

At line 64, the code set the tstk to zero is missing. But in the requesting step (line 17), only tstkp is requested.

Unable to find the following csv: $pathdata/DataRawAndrew/iclink.csv

Am I missing something?

Going through the daily returns data, the portfolios are split sometimes 3, 5, 10, was this due to some constraint?

I had a look the source code offfind comand. And the behavior of ffind has a minor difference from FF48 industry from French data library. ffind groups all sic codes which are not sitting in any of the 47 industries to Other (Almost Nothing) industry.
However, FF48 defines Other industry as below:

4950-4959 
4960-4961 
4970-4971 
4990-4991 

ffind includes the following sic codes in Other industry: 900, 3990, 6797, 9995, 9997. Those sic codes are not in FF48 Other industry.

For your information, two files call ffind command: Stata/14_PrepareAnnualCS.do and Stata/30_CreateSignalsNoFilterVersion.do.

The Piotroski Score at the very bottom shows that for each row there is at least on portfolio value missing (null, NaN)

The PS score is the weirdest of them all:

Originally posted by @firmai in #43 (comment)

The OptionMetrics Manual describes the close field in secprd as "If this field is positive, then it is the closing price for the security on this date. If it is negative, then it is the average of the closing bid and ask prices for the security on this date. In case there are no valid bid or ask for the day, the record does not appear in the table at all." So I think there should be an abs function for close in this script.

Who is going to be the brave soul that turns this into Python code? :)

@velikov-mihail and I updated our HF trading cost code so it just takes an hour to run, provided that one has access to WRDS intraday indicators:

https://github.com/chenandrewy/hf-spreads-all

we should link to this instead of the old code that takes weeks to run.

• needs to be done before task 2

Line 43 in Signals/Code/DataDownloads/I_CRSPmonthly.do: replace ret = ret + dlret.
In this way, 62 observations of delisting adujsted ret are less than -1. The minimum ret is -1.99

Is it better to use replace ret = (1+ret) * (1+dlret) - 1 as suggested by WRDS? There will be no delisting adjusted return that is less than -1.

There should be a sentence 'VOLUME=vol;' in the first block code. Otherwise, VOLUME is uninitialized.

We should just recommend adding these lines:

Signals/Data/**
Shipping/Data/**
Portfolios/Data/**

Otherwise git will crash and you'll have to guess why

In your project you have mainly two datasets needing manual download:

|---- Calls 12_PrepareOtherData.do
|---- Uses pin1983-2001.dat which is currently not being downloaded automatically
|---- Uses SinStocksHong.xlsx which is currently not being downloaded automatically

Could you please indicate from which sources you downloaded the two files and whether they are publicly available?

Thanks for the great project!

I am not able to run WRDS_DL_IBES.sas without including the location of SAS macro:

%include '/wrdslin/wrdsmacros/iclink.sas';

Is there perhaps a dataset where I can pick up the t-stats with minimal effort, to reconstruct the plots as you have in your paper. Thanks in advance.

I am not sure if my understanding here is correct or not. So could you please see my question below? Thanks in advance.

There might be a random order issue about the code block in line 35 to 40 from Signals/Code/DataDownloads/C_CompustatQuarterly.do.

// Prepare year-to-date items
sort gvkey fyearq fqtr
foreach v of varlist sstky prstkcy oancfy fopty {
    gen `v'q = `v' if fqtr == 1
    by gvkey fyearq: replace `v'q = `v' - `v'[_n-1] if fqtr !=1
}

There are some observations with the same gvkey, fyearq and fqtr. In those cases, the order will be random (see page 2 via this link: https://www.stata.com/manuals/dsort.pdf). So the calculation might not always use the most recent information.

When sorting by gvkey, fyearq and fqtr, the following example can happen:

For gvkey 001444 on 30jun2004, oancfyq is -2.579 which is -3.315 (30jun2004)+0.736 (31oct2003).

Below should make sure the order is correct.

sort gvkey fyearq fqtr datadate
foreach v of varlist sstky prstkcy oancfy fopty {
    gen `v'q = `v' if fqtr == 1
    by gvkey fyearq (fqtr datadate): replace `v'q = `v' - `v'[_n-1] if fqtr !=1
}

I found a few errors: AM, BookLeverage, MeanRankRevGrowth should be 12.

This script downloads zips from the BEA and then loads the xlsx files within them. The xlsx filenames seem to change as time goes by (2019 becomes 2020) and ideally we should update elegantly.

A PhD student who would like to remain anonymous writes:

I am writing to confirm the detailed definition of two signals in the SignalDocumention.csv (AddInfo).

1. Accruals (Sloan, 1996).

Detailed Definition: 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.

In Sloan(1996), the two ‘minus’ seem to be ‘plus’, Also, Sloan(1996) substracts depreciation and amortization expense.

2. AnnouncementReturn (Chan, Jegadeesh and Lakonishok, 1996).

Detailed Definition: 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.

The definition in Chan, Jegadeesh and Lakonishok (1996) seem to be ‘from two days before an earnings announcement to one day after the announcement.’

Thank you for this amazing resource.
In the Appendix, it is mentioned that the Size factor (from Banz 1981) is available (it is undoubtedly an important one).
However, I fail to find it under its name (Size) in both the base and additional files.
Has the column name changed?
Many thanks!

Hi Guys

Thank you for a fantastic and fascinating article.

I have a few questions on the signalbase.csv dataset

1. Is permno the Key/UniqueID for every company?
2. Which field did you use for the Liquidity filter?

Kind regards

86ea580

STreversal Daily data compounded returns, turns a dollar into a few billion dollars (all the other anomalies are squeezed to the bottom in a normal range, I am wondering if STreversal, IndRetBig, and IntMom might not have some data error.

Line 326-329
Reads this:

ipos = read_excel(path = tmp) %>% 
  transmute(Founding = Found,
            Offerdate = `Offer date`,
            CRSPperm = PERM)

Should Read:

ipos = read_excel(path = tmp) %>% 
  transmute(Founding = 'Founding',
            Offerdate = 'Offer date',
            CRSPperm = 'CRSP PERM')

WRDS SAS macro keeps the ticker with lower score when using exchange ticker to complete the matching. However, there are four cases as below where the scores are the same.

For the SAS macro: they sort by ticker and score, and then keep the the first ticker. This can filter out ticker with lower score when the scores are different. But this will randomly select the mathcing if the scores are same.

11_PrepareLinkingTables.do requires score not greater than 2. So the only question is the ticker PD2. How to choose the matching since they have same score: further check this one or kick this ticker out?

I do not think this will change the empirical results and just for your information.

ticker   permno  score
PD2      84155       0
PD2      90999       0
permno 84155 was kept from SAS code

KRE1    65162        5
KRE1    65170        5
permno 65170 was kept from SAS code

CMD1  52652         5
CMD1  63933         5
permno 63933 was kept from SAS code

MP      13135       6
MP      55336       6
permno 55336 was kept from SAS code

There was no problem with the code. I just needed to upgrade Rstudio to v1.4.1106. Not sure which version I had before. I removed the gc() commands I sprinkled around and had pushed erroneously. 964a861