Working with Analyst Data: Overview and Empirical Issues

Working with Analyst Data:

Overview and Empirical Issues

Denys GlushkovAssociate Director, WRDS

WRDS Users Meeting

The Wharton School, University of Pennsylvania

October 1, 2007

Outline

Brief overview of IBES database, its coverage and structure

Empirical Issues in Research

IBES Adjusted Data

IBES Unadjusted Data

Relationship between Summary and Detail History

“Rewriting History”

IBES Recommendation History: Linking and Coverage

Tips on working with IBES International

Other

Conclusions

Overview of IBES database

The IBES (Institutional Brokers Estimate System) database

provided by Thomson Financial contains comprehensive global

information on analyst estimates of earnings per share including

both individual forecasts and consensus data. Analyst

recommendations are available as well

In addition to earnings estimates data are available on a more

limited basis for estimates of revenue, cash flow, FFO, EBITDA,

long-term growth and more

IBES Coverage

The IBES database currently covers over 63,000 securities for more

than 50,000 companies in more than 150 industries

Securities in 84 countries

41.5% are for the US including 7% Canadian companies

58.5% are for non-US (international plus some Canadian)

As of Sep 2007, more than 2,500 firms contribute data to I/B/E/S,

from the largest global houses to regional and local brokers, with US

data back to 1976 and international data back to 1987

Basic IBES Structure

IBES

Adjusted

Detail Summary

Unadjusted

Detail

Recommendations

Summary

IBES File Structure

Currently, IBES database consists of 25 files. Main of them are

Detail History and Actuals - DET and ACT (Adjusted), DETU and ACTU

(Unadjusted)

Summary History and Actuals– STATSUM and ACTPSUM (Adjusted),

STATSUMU and ACTPSUMU (Unadjusted)

Recommendations Detail and Summary History – RECDDET and

RECDSUM

Other: ADJ and ADJSUM (adjustment factors), ID, IDSUM, RECDID,

RECIDSUM (Company Identifiers), STOP, EXC, EXCU (Stopped and

Excluded Estimates) and others

Comparing Adjusted and Unadjusted IBES Data

Adjusted and Unadjusted Data line up well:

DET and DETU (only for EPS measure): only 0.01% of extra records in

adjusted file, mainly for 1986-1993

STATSUM and STATSUMU (only for EPS and EPSPAR measures):

27,295 obs in the Adjusted Summary File, which are not in Unadjusted

Summary - roughly 0.2% of all Adjusted Summary records.

All of these differences are for International firms only

Relationship between First Call and IBES

The Populations of broker/analysts are likely to differ between the two

databases

Brokers are not required to submit their analysts’ forecasts to these

organizations

Variables and time series are likely to differ as well

First Call contains company guidelines, but IBES contains more forecast

variables and a longer time series

Linking both databases is quite complicated

IBES Cusip is historical, FC Cusip is the header (most current)

Broker IDs and Codes are different for both Estimates and Recommendations

E.g., AG Edwards has Broker ID 244 in FC Estimates and Broker Code 000094 in IBES

Detail

E.g., AG Edwards has Broker Code “AG” in FC Recommendations and Broker ID

“EDWARDS” in IBES Recommendations

IBES Adjusted Data: Things to know

Traditionally, IBES provided forecast data on an adjusted basis, rounded

to 2 decimal places on the Summary files and to 4 decimals on the Detail

files.

Adjustment and the corresponding rounding in IBES carries over the

entire time-series for a given security

Rounding issue is more severe in cases when

Summary files are used

IBES suggests a zero forecast error since the researcher can’t “unadjust” the data

Year 1

After a 4-

for-1

Stock

Split in

Year 2 *

Company A Earnings 0.99 0.25

Forecast 1.00 0.25

Forecast error -0.01 0.00

Company B Earnings 1.01 0.25

Forecast 1.00 0.25

Forecast error 0.01 0.00

*From "The Implications of Using Stock-Split Adjusted IBES Data in Empirical Research" by Payne and Thomas (2003)

IBES Adjusted Data: Things to know

The adjustment issue becomes more pronounced as the split factor

increases:

Quick check in CRSP suggests that since Jan 1975 there are 517 stocks

which experienced 16-for-1 splits, 213 stocks with 32-for-1 splits and

66 stocks with 64-for-1 splits

Year 1

After a 64-

for-1

Stock

Split in

Year N *

Company A Earnings 0.33 0.01

Forecast 0.64 0.01

Forecast error -0.31 0.00

Company B Earnings 0.95 0.01

Forecast 0.64 0.01

Forecast error 0.31 0.00

*From "The Implications of Using Stock-Split Adjusted IBES Data in Empirical Research" by Payne and Thomas (2003)

IBES Adjusted and Unadjusted data: Things to know

Payne and Thomas (2003) find that research conclusions are more

likely to be affected by the rounding procedure in samples that have

stock splits

e.g., larger firms, higher M/B, better performers, etc

AND

Where the research question focuses on zero forecast error amounts

assessing the percentage of zero forecast errors over time

relating firm characteristics to the probability of zero forecast errors over

time

calculating the market’s reaction to zero forecast error

inferring earnings management based on the distribution of earnings,

earnings changes and forecast errors around zero

What is a researcher to do?

It is useful to know the actual historical amounts that are NOT

adjusted for subsequent stock splits – IBES Unadjusted Data - and

create their own split-adjusted data without rounding to the nearest

penny

If unadjusted data can not be obtained, an alternative is to

recalculate IBES consensus statistics using the detail IBES adjusted

data (which has rounding to 4 decimals)

IBES Unadjusted Data I: Things to Know

Unadjusted data consists of Detail history – Estimates, Actuals,

Excluded and Summary Stats. Contains only EPS-related data

Merging Unadjusted Actuals with Detail History is problematic

If a split occurs between analyst’s Estimate date and the associated Report

Date, the estimates and actual values may be based on different number of

shares outstanding. E.g. Amazon split 2-for-1 on Jun 2, 98 and 3-for-1 on Jan

5, 1999.

TICKER ESTDATS BROKER ANALYS VALUE FPEDATS REVDATS REPDATS Actual

AMZN 20-May-98 86 42186 -2.5 31-Dec-98 03-Jun-98 26-Jan-99 -0.517

AMZN 26-May-98 192 1830 -2.1 31-Dec-98 23-Jul-98 26-Jan-99 -0.517


AMZN 04-Jun-98 86 42186 -1.17 31-Dec-98 22-Jul-98 26-Jan-99 -0.517


AMZN 08-Jun-98 191 45029 -1.21 31-Dec-98 09-Jun-98 26-Jan-99 -0.517

……………

AMZN 05-Jan-99 241 259 -0.67 31-Dec-98 23-Jun-00 26-Jan-99 -0.517




IBES Unadjusted Data II: Things to Know

Potential solutions – IBES Method

Based on adjusting actual values using IBES adjustment factor valid as of

the report date and then unadjusting this adjusted actual using IBES

adjustment factor valid as of the estimate date

The problem is that IBES effective split date is NOT necessarily the true date

of the stock split. Instead, this is the date when the split became “effective”

within the IBES database. IBES “Split” Dates are Jun 18, 1998 and Jan 14,

1999.TICKER ESTDATS BROKER ANALYS VALUE FPEDATS REVDATS REPDATS Actual


AMZN 26-May-98 192 1830 -2.1 31-Dec-98 23-Jul-98 26-Jan-99 -3.102




AMZN 08-Jun-98 191 45029 -1.21 31-Dec-98 09-Jun-98 26-Jan-99 -3.102

AMZN 05-Jan-99 241 259 -0.67 31-Dec-98 6/23/2000 26-Jan-99 -1.551




…………………………………………………………………………………………………………………………

IBES Unadjusted Data III: Things to Know

The same technique is more effective when used with IBES

Unadjusted Summary

IBES Effective Split date is linked to Summary file’s statistical period date,

and both are mid-month dates

Although tests show that results of the merge are significantly better, it is still

possible to find exceptions

Microsoft split 2-for-1 on Apr 12, 1990 and they reported earnings on Apr 17,

1990, whereas IBES effective split date was April 19

So the problem occurs in Summary file whenever Split Date<Report

Date<IBES Effective Split date

Ticker FPEDATS STATPERS REPDATS MEAN ACTUAL

MSFT 31-Dec-89 14-Dec-89 18-Jan-90 1.00 1.26

MSFT 31-Dec-89 18-Jan-90 18-Jan-90 1.01 1.26

MSFT 31-Mar-90 15-Feb-90 17-Apr-90 1.16 0.62

MSFT 31-Mar-90 15-Mar-90 17-Apr-90 1.16 0.62

MSFT 30-Jun-90 19-Apr-90 25-Jul-90 0.59 0.65

MSFT 30-Jun-90 17-May-90 25-Jul-90 0.62 0.65

IBES Unadjusted Data IV: Things to Know

Alternative Solutions:

Method 2: To get the correct split factor, one can merge the unadjusted data

with the adjusted data and then back out the split factor. However,

discrepancies may still occur due to the rounding effects in IBES Adjusted

Data

Method 3: Perhaps, the most accurate (and relatively time-consuming) way

of joining Unadjusted Detail History with Unadjusted Actuals is to use

CRSP cumulative adjustment factor. This involves using the sample code

developed by WRDS which creates a linking table for IBES and CRSP

Details of all three methods as well as the sample SAS code can be

found in the document entitled “A Note on IBES Unadjusted Data”

on the IBES Documentation page

Recreating IBES Summary Stats based on Detail History

IBES includes only effective estimates while calculating the summary

stats from detail, but provides no clear definition of what is considered

an effective estimate

Potential algorithm:

Each forecast in Detail file that is included in the sample used to calculate

NUMEST must:

a. pertain to the same firm, measure and forecast period

b. be issued prior to the Summary record’s statistical period date

c. not be voided by IBES with an “Excluded” or “Stopped Coverage” flag

d. be the latest estimate issued by that particular brokerage once conditions (a)-

(c) above are satisfied

Analysis indicates that IBES polls brokerages, “lumping” together

forecasts of different analysts, so that a forecast made by any one of

them is discarded should any colleague issue one thereafter

Recreating IBES Summary Stats: Illustrative Example

Consider records from Summary and Detail History Files for basic EPS

for Bio Technologies, Inc. for fiscal year 2004

TICKER MEASURE FPEDATS FPI STATPERS NUMEST MEANEST STDEV ESTFLAG

BIT EPS 31-Dec-04 1 14-Oct-04 2 0.14 0.01 P

TICKER MEASURE FPEDATS FPI ESTDATS BROKER ANALYST VALUE

BIT EPS 31-Dec-04 1 9-Feb-04 2192 9479 0.14 superseded

BIT EPS 31-Dec-04 1 7-May-04 2192 72066 0.15

BIT EPS 31-Dec-04 1 3-Aug-04 1996 107152 0.14

BIT EPS 31-Dec-04 2 8-Jul-03 1876 7646 0.23 Stopped

BIT EPS 31-Dec-04 2 9-Oct-03 1876 7646 0.19 Stopped

BIT EPS 31-Dec-04 2 10-Dec-03 1876 7646 0.22 Excluded

BIT EPS 31-Dec-04 8 3-Aug-04 1996 107152 0.04 discard

BIT EPS 31-Dec-04 9 9-Feb-04 2192 9479 0.04 discard

TICKER MEASURE FPEDATS FPI ESTDATS BROKER ANALYS VALUE EXCDATS

BIT EPS 31-Dec-04 2 10-Dec-03 1876 7646 0.22 16-Apr-04

TICKER MEASURE FPEDATS PDICITY BROKER ESTPDATS

BIT EPS 31-Dec-04 A 1876 20-Apr-04

IBES Detail History File Records

IBES Summary Statistics Records

IBES Excluded File

IBES Stopped File

Recreating IBES Summary Stats: Conclusions

How well does the described method match NUMEST for a realistic

number of Summary records?

(a), (b) (a), (b), (d) (a) - (d)

NUMEST matched (%) 0 0 33.3

NUMEST overestimated (%) 100 100 64.2

NUMEST underestimated (%) 0 0 2.5

Average NUMEST error 651.1 34.6 5.1

Average MEANEST error 7.7 2.1 0.4

Average STDEV error 467 309.7 37.8

Cases MEANEST error < 5% (%) 60 83.3 99.2

Cases STDEV error < 5% (%) 0 1.7 33.3

(a), (b) (a), (b), (d) (a) - (d)

NUMEST matched (%) 3 28 80

NUMEST overestimated (%) 97 71 19

NUMEST underestimated (%) 0 0 2

Average NUMEST error 642 54 7

Average MEANEST error 49 20 4

Average STDEV error 509 311 41

Cases MEANEST error < 5% (%) 3 29 81

Cases STDEV error < 5% (%) 31 67 92

Panel A: Case 1 (IBM only)

Filter rules applied

Panel B: Case 2 (All firms, fiscal year ending Dec 2004)

Rewriting History

Ljungqvist, Malloy and Marston (2007) compared two snapshots of

IBES recommendations database (taken in Sep 2002 and May 2004) for

the same time period Oct 29, 1993-July 18, 2002 and find in a dataset of

280,464 obs there are 54,729 non-trivial ex-post changes. Of them

19.5% Alterations – records whose recommendation levels are different on the

2002 and 2004 tapes

9% Deletions - records that appear on the 2002 tape but not on the 2004 tape

35.1% Additions – records that appear on the 2004 tape but not on the 2002 tape

36.4% of anonymizations – records attributed to specific analysts on the 2002

tape whose names on the 2004 tape (and whose amaskcd identifiers have been

set to zero)

Thomson Financial have recently reinstated the missing analyst names in

the recommendation history file – as of Feb 12, 2007, the data on WRDS

now reflect this reinstatement

IBES Vintage Changes, part I

According to the informatiom from Thomson Financial

“the great majority of the records missing in the July 2007

vintage are for brokers Merrill Lynch (non-US and Canada)

and Lehman Brothers (Europe and Global), due to requests

from the two brokers that WRDS does not have access to their

forecast data”

Recommendation History, however, does not appear to have the same issue

WRDS identified 57 unique broker ID which were present in Dec

2002 vintage of Detail History, but no longer exist in July 2007

vintage.

Majority of these “dissappeared” brokers are international such as JP

Morgan (Intl), Bear Stearns &Co Inc (Latin), Morgan Stanley Pacific

Basin and ABN Amro Australia Ltd

IBES Vintage Changes, part II

Though IBES actual values are “as reported” and not “restated”, but

we found some observations for which actuals were changed

between the vintages (march 2007 and aug 2007 were used):

For USFIRM=1 and EPS measure only there is 1.7% of changed actuals. 88%

of them have differences more than 5% in absolute magnitude

For USFIRM=1 and all measures, there is 0.8% of changed actuals. 67% of

them have differences more than 5% in absolute magnitude

For USFIRM=0 and EPS measure, there is 3% of changed actuals

There does not seem to be a drastic number of changes in actuals

and these findings are likely to reflect regular revisions/corrections

IBES does to the database

Linking Detail History and Recommendations

Matching should be performed by the analyst codes in the two files,

ANALYS and AMASKCD.

The analyst codes are the same in both files, but the broker codes are not (i.e., a

given broker may have different codes in the two files).

For each Detail History EPS estimate, match all unique

recommendations that existed during the period that the estimate

was still valid. That is,

If a given analyst estimate was in effect between ESTDATS and REVDATS,

each unique recommendation that existed for at least one day within that time

period will generate a new record in the merged file

The first date of the recommendation must come on or before the last date of

the estimate, and the last date of the recommendation must come on or after the

first date of the estimate.

Linking Detail History and Recommendations

Timeline:

ESTDATS RECDATS EST.REVDATS REC.RREVDATS

ORRECDATS ESTDATS REC.RREVDATS EST.REVDATS

PROC SQL; CREATE TABLE final

AS SELECT det.*, recddet.bmaskcd , recddet.recdats,

recddet.rrevdats, recddet.brokerid, recddet.itext

FROM detsamp AS d LEFT JOIN recsamp AS r

ON (d.ticker = r.ticker and d.analys = r.amaskcd)

AND (r.recdats <= d.revdats and r.rrevdats >= d.estdats)

QUIT;

Tips on International Aspects of IBES I

CUSIP in IBES could represent either a CUSIP or SEDOL depending

on whether the company is foreign or not

Country variable is not directly available, but can be identified by the first two

digits of CUSIP variable

USFIRM dummy variable refers to the firm being followed, not the

broker or analyst

According to IBES, “international companies (i.e. non-US companies)

are covered by international (non-US) analysts and US companies are

covered by US analysts

If the broker of the analyst which covers an international firm is a US

company, the analyst is located in a foreign office of that broker

Tips on International Aspects of IBES II

Canadian firms are considered both US firm and an “International firm”

in IBES Detail History. This results in duplicate records for each

Canadian company

IBES uses the convention of putting an “@” symbol in front of tickers of

international companies, with the exception of Canadian firms

Hence, if USFIRM=0 and ticker does not have “@” in it, this is a Canadian

company

Currency issues

CURRFL in the Detail History indicates the currency in which the estimate was issued

for US/Canadian companies. It is missing if the estimate is in the US dollars and takes

“C” if the estimate is in Canadian dollars. For international companies, it indicates

whether the estimate was issued on a parent (“P”) or consolidated level (“C”)

CURR in the Detail History is missing for US/Canadian firms and indicates the

currency in which the estimate was issued for international companies

CCOPCF in the Company Identification file has the same meaning as CURRFL, only

at the company, not estimate, level

CURCODE in the Summary History is the same as CURR, but at the company level

Difference in IBES and Compustat actual EPS numbers

Livnat and Mendenhall (Journal of Accounting Research, Mar 2006)

provide a nice summary for why differences may arise:

Compustat follows a policy of changing firms’ recorded quarterly earnings figures

to reflect restated values. IBES includes the originally reported earnings

Compustat’s earnings reflect GAAP, while most analyst-tracking services report

“street” measures of earnings. That is, IBES earnings exclude various expenses

required by GAAP. Some expenses omitted from street earnings can be found in

Compustat’s variable “Special Items” and can be backed out.

IBES observes the market reaction to the earnings announcement prior to choosing

exactly which earnings components to include in street earnings. This leads to a

potential ex post selection bias.

IBES uses weighted average of the shares outstanding in EPS calculations, but does

not report this average

Also a great overview of the issue is in “GAAP vs. The Street: an

Empirical Assessment of Two Alternative Definitions of Earnings” by

Bradshaw and Sloan (Journal of Accounting Research, 2002)

Broker ID, Broker Masked Code and Broker Translation

Be aware that some brokerage firms may have several broker ID –

for example if they have offices around the globe:

Thomson no longer provides either the Broker and Analyst

Translation (often referred to as BRAN) or the Recommendation

Broker Translation files (often referred to as RECDBKR).

Broker ID BMASKCD

BARINGLB 00002689

BRNGEEUR 00002689

FERRI 00002689

FURMAN 00002689

ING 00002689

Other issues in IBES

There are estimates which are dated “after” the announcement date. We

have no explanation other than the entry is in error. % of such

observations in the latest vintage (Aug 2007) is very small (0.04%)

IBES defines REVDATS as ”most recent date that an estimate was

confirmed as accurate”. Note that

Announcement of earnings will increment the FPI variable by 1 in all IBES

records for which review date (REVDATS)> report date (REPDATS)

If at the time of the next review date the same analyst at the same brokerage

changes her forecast for the same (TICKER, ESTDATS, FPEDATS, FPI,

MEASURE, USFIRM) combination, IBES will add a new observation. If the

forecast remains unchanged, IBES will not add new observations, but will

adjust the review date (REVDATS) accordingly

Conclusions

Working with IBES requires good understanding of some issues:

Be aware of rounding issues in Adjusted Data which may lead to wrong

estimates of earnings surprises

Make sure to link Unadjusted Data and Actuals so that estimate and actual

are on the same per-share basis (see more on IBES Documentation page)

Replicating Summary from Detail appears to work better when the latest

estimate for a given broker (not analyst) is included

Make sure to understand the difference between Currency Flags, Indicators

and Codes in IBES International data

Understand the difference between GAAP earnings (Compustat) and Street

earnings (IBES, First Call)

Due to unavoidable changes and revisions IBES makes from one vintage to

another, WRDS recommends working with a saved vintage of IBES

References

1. Bradshaw, M. and Sloan, R. (2002). ”GAAP vs. The Street: an Empirical Assessment

of Two Alternative Definitions of Earnings”. Journal of Accounting Research, Vol. 40

No. 1, pp. 41-66

2. Glushkov, D. and Robinson D. (2006). “A Note on IBES Unadjusted Data”, WRDS

Documentation on IBES, available at

http://wrds.wharton.upenn.edu/ds/ibes/lib/IBES_Unadjusted_Data.pdf

3. Ljungqvist A., Malloy, C. and Marston F. (2007). “Rewriting History”, working paper,

available at http://papers.ssrn.com/sol3/papers.cfm?abstract_id=922053

4. Livnat, J. and Mendenhall, R. (2006). “Comparing the Post-Earnings Announcement

Drift for Surprises Calculated from Analyst and Time Series Forecasts”. Journal of

Accounting Research, Vol. 44 No. 1.

5. Payne, J. and Thomas B. (2003). “The Implications of Using Stock-Split Adjusted

I/B/E/S Data in Empirical Research”. Accounting Review, 78(4), 1049-1067

6. Shvorob, D. and Vora, P. (2006). “Recreating IBES Summary based on Detail History

File”, WRDS Documentation on IBES, available at

http://wrds.wharton.upenn.edu/ds/ibes/lib/IBES_Summary_from_Detail.pdf

http://wrds.wharton.upenn.edu/ds/ibes/lib/IBES_Unadjusted_Data.pdf

http://papers.ssrn.com/sol3/papers.cfm?abstract_id=922053

http://wrds.wharton.upenn.edu/ds/ibes/lib/IBES_Summary_from_Detail.pdf

Working with Analyst Data: Overview and Empirical Issues

Documents

Transcript of Working with Analyst Data: Overview and Empirical Issues