Why the estimation period matters even with MAR

Users sometimes ask why Eventus seems to expect an estimation with market-adjusted returns (MAR). Sometimes users experiment with different estimation periods with MAR and observe changing results, which can seem unexplainable.

With a MAR event study, Eventus uses the estimation period to calculate standard deviations, ranks and sign percentages, so most test statistics are at least a little sensitive to the estimation period. Sample characteristics can determine that returns differ greatly between two estimation periods, increasing the variation in test statistics in such cases.

Also, the mean or median CAR can change because changes in estimation period settings alter the sample composition. For example, if we set MinEstN=10 and experiment with EstLen=100 versus EstLen=10, we may observe different sample compositions. With EstLen=10, MinEstN=10 excludes any security-event that has a single missing return in this 10-day period, whereas with EstLen=100, such a security-event would not be excluded as long as it had no more than 90 missing return days in the 100-day period.

Changing the Est= option can also affect missing value screens, and even can move the estimation period to a date range during which a particular security is not listed.

Users asking about the latter situation often submit code including the NoNames or NoPrint options. The first sections of the default Eventus output (in the Ouptut window or *.lst file) are designed to allow the user to see which security-events are included or excluded. It makes sense to suppress this output when one is already aware of and satisfied with the sample that survives, but allowing it to appear (by omitting NoNames and NoPrint) can provide useful information when one is still testing and developing the Eventus code for a particular research need.

If you aren't interested in standard deviations or test statistics, you can do the following to minimize the impact of the estimation period: Add EstLen=10 Est=+1 (or -1) to the Request statement and Overlap to the EvtStudy statement. This will overlay the estimation period on the event period.