How to calculate Forecast Accuracy in IBES: monthly EPS estimates (forecasts) versus annual actual EPS. 
Posted: 25 March 2016 07:58 AM   [ Ignore ]
Newbie
Rank
Total Posts:  5
Joined  2016-03-23

Hi everybody,

I am trying to calculate forecast accuracy using the IBES database.
The exact calculation is as follows: absolute value of (“the median of one year ahead analysts’ forecasts for EPS for period t ” minus “actual EPS in period t”) divided by “price per share at the beginning of period t.”

Now I am a bit confused as how to do this in IBES. This is what I have so far:

• Median of one year ahead analysts’ forecasts of EPS.
I downloaded the IBES Unadjusted Summary file (FPI = 2, since I need one year ahead estimates): this gives me monthly median EPS estimates. I think alternatively I could also download the unadjusted detail file and then calculate the median estimates per month, but I guess this should be approximately the same.

• Actual EPS:
Here I use FPI = 1. I downloaded the IBES Adjusted Summary file with the data item “actuals from details files” (the unadjusted summary file does not have this actual data item). These actuals seem to match with the results of IBES Detail Unadjusted Actuals file, which I also downloaded. So far so good.


My problem =  the periods do not match.
Actual EPS: these are given on an annual basis (is this EPS at fiscal year end?).
Forecasted median EPS: monthly median estimates (in Summary file), or monthly detailed estimates (from Details file).

What should I do? Take the average of the 12 monthly median estimates from the Summary file to come to an “average” median estimate per fiscal year & then compare this with the annual actual EPS? I am not sure how else to fix it.

Moreover, I am not sure where to get the price per share at the beginning of period t. It seems that I cannot get this out of IBES. Can I perhaps use the pccc-f (closing price, fiscal) from Compustat and then match on the basis of 9 digit CUSIPS?


Any help is very much appreciated! Thank you in advance for your time and effort.

Best regards,

Elisa

Profile
 
 
Posted: 25 March 2016 08:07 AM   [ Ignore ]   [ # 1 ]
Administrator
Avatar
RankRankRankRank
Total Posts:  901
Joined  2011-09-19

hi Elisa,

It sounds like you haven’t decided when to measure the analyst forecast. Say some fiscal year is jan 1 - dec 31 2015. Earnings for Q4 will be reported around Feb-March 2016 (and with Q4, the full 2015 earnings will be known).

There is the ‘walkdown’ to beatable earnings paper (http://onlinelibrary.wiley.com/doi/10.1506/KHNW-PJYL-ADUB-0RP6/abstract ) that shows that there is analyst optimism that slowly disappears over time. For example, forecasts given in 2013 for fiscal 2015 are more optimistic compared with forecasts given in 2014, 2015 etc. Right before earnings announcements the forecasts are most realistic. So, generally it will matter when you measure the forecasts (likely to be differences between firms with respect to the optimism—analysts may be more optimistic to certain firms, like high growth firms etc).

If you are following some paper, it should be explained when the forecasts were measured. If not, the conservative thing to do is to construct your variable in several ways (e.g., measure first month of fiscal year, measure halfway, measure last month, etc).

prcc_f in Funda would work for stock price.

linking Compustat with IBES is not easy—see this page (check link to iclink.sas)—http://www.wrds.us/index.php/forum_wrds/viewthread/467/

Hope this helps,

Joost

 Signature 

To reply/post new questions: Please use the group WRDS/SAS on Google Groups! http://groups.google.com/d/forum/wrdssas

Profile
 
 
Posted: 25 March 2016 08:54 AM   [ Ignore ]   [ # 2 ]
Newbie
Rank
Total Posts:  5
Joined  2016-03-23

Hi Joost,

Thanks for your fast response! It definitely helps.

I looked it up in the paper that I am following, and they say that: “since they are interested in assessing the impact of disclosures generally, rather than at a particular announcement date (...) and forecast accuracy are computed as the simple average of the measure across 12 monthly reporting periods on the IBES tape during the company’s fiscal year.” So it seems that they take the average over 12 months then.

As with regard to the match with IBES. I have both gvkeys and 8 digit CUSIPs available for my sample. What I did now, was to take the 8-digit (which I used for IBES), put those into the WRDS CUSIP conversion tool and through that tool I obtained the associated 9 digit CUSIPS. In Compustat one of the outputs that I have is the 9 digit CUSIP. So that is how I matched the two databases: by using the 9 digits CUSIPs. Is this ok? Or do I really need to use the CCM linking table to get Permno’s (I did this already as well), and then use a code to match it with IBES? The only thing is that I have no experience with SAS whatsoever, so I was hoping that the CUSIP way would be fine. I am using big US firms by the way. smile

Thank you very much for you help!

Elisa

Profile
 
 
Posted: 25 March 2016 10:39 AM   [ Ignore ]   [ # 3 ]
Administrator
Avatar
RankRankRankRank
Total Posts:  901
Joined  2011-09-19

hi Elisa,

You’re welcome!

Which % of your observations did you match with Cusip?

Best,

Joost

 Signature 

To reply/post new questions: Please use the group WRDS/SAS on Google Groups! http://groups.google.com/d/forum/wrdssas

Profile
 
 
Posted: 25 March 2016 11:17 AM   [ Ignore ]   [ # 4 ]
Newbie
Rank
Total Posts:  5
Joined  2016-03-23

Hi Joost,

I matched 64%, is that ok?

Best regards,

Elisa

Profile