Point-in-time or look-ahead bias?

When I retrieve data through Python API by referring to fields such as TR.PriceToSalesPerShare or TR.F.ReturnAvgComEqPctTTM getting both date and value, is it point-in-time data? That is if the data is retrieved e.g. for 2019-12-31 does it take into account only the reported financial data (like sales, earnings or equity) that had been available for this date? Or is there a look-ahead bias present?

Find more posts tagged with

python

eikon

time-series

fundamentals

workspace

eikon-data-api

refinitiv-dataplatform-eikon

workspace-data-api

Accepted answers

Alex Putkov.1

I'm afraid it's very hard for me to make generalized statements while I'm not sure I'm following your logic or thought process. The example of a potential for forward looking bias that you gave for TR.PriceToSalesPerShare does not apply to TR.F.ReturnAvgComEqPctTTM, which is calculated solely from items in company reports and where I don't see any potential for mismatching time periods. All the components of TR.F.ReturnAvgComEqPctTTM ratio become known at the same time (i.e. on the date of quarterly earnings release), and this is when the ratio is calculated. What I'm guessing you may be asking about is how ratios like TR.F.ReturnAvgComEqPctTTM are timestamped. Or, in other words, what do TR.F.ReturnAvgComEqPctTTM.date and TR.F.ReturnAvgComEqPctTTM.calcdate represent when you retrieve timeseries for TR.F.ReturnAvgComEqPctTTM. For ratios calculated solely from items in company reports ".date" represents the end date of the fiscal period, for which the ratio is calculated. ".calcdate" refers to the date when the calculation is performed. Using as an example

ek.get_data("IBM.N",
            ['TR.F.ReturnAvgComEqPctTTM.date',
             'TR.F.ReturnAvgComEqPctTTM.calcdate',
             'TR.F.ReturnAvgComEqPctTTM'],
            {'SDate':'2021-01-01', 'EDate':'2021-01-31', 'FRQ':'D'})

you can see that for calculation dates before the earnings release date of 2021-01-21 the corresponding ".date" returns 2020-09-30, and for calc dates after 2021-01-21 the corresponding ".date" returns 2020-12-31. You also see that in daily timeseries the value of the ratio calculated on 2021-01-21 is repeated for subsequent calc dates until the next earnings release date. I suppose in your definition this means the ratio has forward looking bias if you anchor it on ".date", but doesn't if you anchor it on ".calcdate". Am I correct?
Does this help? Does this give you the answer you're looking for?

All comments

Alex Putkov.1

@jakub.wielgus

You can use CodeCreator app in Eikon or Refinitiv Workspace to lookup the definitions of any fields you can retrieve using Eikon Data APIs. Here are the definitions for the two fields you mentioned:

TR.PriceToSalesPerShare
Price To Sales Per Share is calculated by dividing the company's market capitalization by its total sales. Market Cap is calculated by multiplying Current Total Shares Outstanding by Latest Close Price. Sales is LTM Total Revenue. For Banks, Interest Income plus Non-lnterest Income is considered as Total Sales. Price to Sales Per Share is not calculated when LTM Sales is less than or equal to Zero.
Category: Historical Time Series Ratios

TR.F.ReturnAvgComEqPctTTM
Return on Average Common Equity - % (Income available to Common excluding Extraordinary Items), TTM [RetOnSTCSE_TTM] represents Income available to Common excluding Extraordinary Items [TR.F.lncAvailToComExclExordltems (Methodology=StubLtm)] / Shareholders Equity - Common, YoY Avg [TR.F.ShHoldEqComYoYAvg] * 100. Denominator should be positive. If numerator and denominator are both zero, this will return a value of zero. The data item is calculated for Interim periodicities only. It is applicable to all industries.
Category: Cross Statement - Profitability

Based on these field definitions I don't see how any forward looking bias could be factored into the values.

jakub.wielgus

Thank you. I looked at those definitions, but for me it is not 100% clear, whether there is no forward looking bias. They do not indicate the relationship between variables used to calculate those ratios and the availability of data on those variables. Is it possible to get 100% confirmation that only point-in-time data is used to calculate them?

Alex Putkov.1

@jakub.wielgus

I'm afraid I'm not following where your doubts come from. Could you please elaborate? Where in these descriptions do you see any potential for forward looking bias? It would be really helpful to get an example of how in your opinion forward looking bias could potentially factor into these ratios given the definition provided. Otherwise, I'm afraid it's hard for me to give you a 100% confirmation for something I'm not quite following. I start suspecting that maybe you and I have different understanding of what forward looking bias means? For me forward looking bias occurs when a measure factors in some kind of forecast of future events.
From my perspective, taking the example of Price To Sales Per Share, none of the components in the calculation could possibly be forward looking. According to the definition the components of the ratio are:

Current Total Shares Outstanding
Latest Close Price
LTM (last 12 months) Total Revenue

Current Total Shares Outstanding can't be forward looking, right? Latest Close Price can't be forward looking either. It's not a price forecast, it's the latest historical close price published by exchange. LTM Revenue also can't be forward looking. It's the last 12 months revenue, not next 12 months revenue forecast or some such.
Could you please explain where you see the possibility for forward looking bias?

jakub.wielgus

In this case, LTM Revenue could be forward looking. Today the LTM Revenue for any given company is not known, it is being reported in certain (mostly quarterly intervals). So for company X today I might know LTM Revenue but counting not from today but e.g. from 2020-12-31. Point-in-time data means that for a given day only date that was publicly available on this date is shown. Forward-looking bias means that the data for that date is subsequently updated and e.g. in half a year when analysing data as of 2021-02-28 I would be shown revenues for 12 months before this date even though this information was not publicly available on this very date.

Does the above explanation clarify the source of my doubts?

Alex Putkov.1

Thank you for the explanation. I see what you mean now.

It's quite easy to verify that calculations are performed using the data available as of the date associated with the value in the timeseries. Consider this example.

On 15-Jan-2021 market cap for IBM was $114,402,823,123.239 ("TR.CompanyMarketCap(SDate=20210115)").

Total revenue for 4 quarters ending on 30-Sep-2020 ("SUM(TR.Revenue(Period=4FQ2019:3FQ2020))") is $75,031,000,000, which gives the Price to Sales per Share ratio of 1.52474074880036 exactly corresponding to the value returned by "TR.PriceToSalesPerShare(SDate=20210115)".

If the calculation was performed using total revenue for 4 quarters ending on 31-Dec-2020 ($73,621,000,000), which wasn't known on 15-Jan-2021, the ratio would have been calculated as 1.553942803.

jakub.wielgus

Thank you for this confirmation regarding TR.PriceToSalesPerShare which I take to be applicable also to other indicators from Daily Time Time Series category. Could you please confirm that this is also the case for indicators from TR.F branch, e.g. the one I mentioned TR.F.ReturnAvgComEqPctTTM ?

Alex Putkov.1

ek.get_data("IBM.N",
            ['TR.F.ReturnAvgComEqPctTTM.date',
             'TR.F.ReturnAvgComEqPctTTM.calcdate',
             'TR.F.ReturnAvgComEqPctTTM'],
            {'SDate':'2021-01-01', 'EDate':'2021-01-31', 'FRQ':'D'})

EXPLORE OUR SITES