|
| |
| Patrick De lamirande and Jason Stevens |
| |
| ''Predicting events with an unidentified time horizon'' |
| ( 2016, Vol. 36 No.2 ) |
| |
| |
| Economists often employ binary choice models to determine if variables of interest such as asset prices or returns are able to predict the occurrence of significant events, most notably recessions.
It is, however, unclear how the results of existing studies should be interpreted due to the common practice of testing the predictability of the event at multiple horizons. Presented with a set of test statistics, some may be tempted to conclude that the variable of interest is able to predict the event if the null hypothesis of non-predictability is rejected at any horizon. This paper demonstrates that this approach results in a significant probability of spuriously concluding that the event of interest is predictable.
In light of this possibility, the ability of the term spread to predict US recessions is re-examined with corrected critical values, confirming that the results found in the existing literature are not the result of data-snooping.
|
| |
| |
| Keywords: Spurious regressions, predictability, binary choice models. |
JEL: C5 - Econometric Modeling: General
|
| |
| Manuscript Received : Sep 19 2014 | | Manuscript Accepted : Apr 14 2016 |
|
