A horse of a different statistical color

There's been a lot of talk about the sales numbers coming out of the Keeneland yearling sale.  We've heard double-digit percentage gains thrown about by the media and Keeneland itself.

It's heartening to hear these numbers, since we all want to see the racing industry bounce back from the recent crushing economic blows.  But are the numbers accurate?

It's fair to question them since Keeneland changed its sales format this year but hasn't changed its method of comparison -- that is, for example, comparing Day 1, 2009 to Day 1, 2010.  If there are fewer horses being offered the first two days of the sale this year, it seems there should be an adjustment in the way the two years are compared.

We've been talking about this quite a bit in my equine finance course at the University of Louisville, and our professor has come up with an alternative method.  He's comparing the sales in terms of percentage of horses catalogued.  It makes sense.

I can't say with 100% certainty that his method is without flaw or that the numbers are 100% accurate (although I'm confident that they are), but it gives us a different way to measure the Keeneland sale in any case.  It's important to look at these things with a healthy dose of realism, even though we might want to see eye-popping numbers.   So, without further adieu, here's professor Robert Losey's explanation:

Keeneland September Sales Results: Are They Really Up That Much?

Making sense of what is happening to prices at the Keeneland September sale is especially difficult in 2010.   There have always been problems in comparing one year to the next:  One complication relates to the length of sales.  In 2008 Keeneland September had 15 sale sessions: In September 2009 it had only 14.  Reports we have seen compared the first day 2009 to the first day 2008 and continued the comparison out until they compared the 14^th (which was the final) day 2009 to the 14^th day 2008.  This left the 15^th (final) day 2008 in a state of limbo where it was not compared to anything.  More reasonably, it would have made sense to compare the two final days (14^th day 2009 and 15^th day 2008) to each other, as each had similar quality horses.  

This year, comparisons have become much more difficult because of Keeneland’s new format.  Should the 105 horses from day 1 2010 be compared to the 207 horses offered on day 1 last year?  Should day 6 this year, which is still part of the select portion of the sale, be compared to last year’s non-select day 6, or to day 4 in 2009, the final select day last year?  

Kate Ellis, an equine certificate student in the Equine Industry Program at the University of Louisville, working with Robert Losey, Professor of Equine Management, has devised a statistical “fix” that makes comparisons of sale averages from one year to the next more logical.  Rather than report day by day comparisons, the Ellis approach compares the day 1 sales this year (which made up the first 2.16% of the total yearlings offered) to the sales average for the first 2.16% of horses from last year’s sale.  Subsequent comparisons follow similar lines, always comparing horses offered in comparable percentage segments of each year’s sale.  This approach has the salutary effect that it is more likely to compare horses of similar quality:  Early-sale horses are compared, mid-sale horses are compared, and end-of-sale horses are compared.

Because of the change in the format for the 2010 sale, which featured very small first and second-day “super-select” yearlings, our method, though an improvement over what is typically being reported, still provides an apples-to-cranapples comparison on a day-by-day basis for the first six sessions for 2010.  However, with the completion of the sixth session of Keeneland on Friday, our method provides an apples-to-apples comparison for the cumulative average starting with day 6 and for each subsequent session.  Going forward our method also provides a useful basis for comparing each 2010 sale day to a comparable segment of last year’s sale.  

Compare our method to reports from other sources.  For the second Monday (Sept. 20, 2010), other sources compared this session, which was the eighth session, but the second non-select session in 2010, to the eighth session 2009, which was the fourth non-select session in 2009.  Not surprisingly, because Keeneland tries to place yearlings with the highest values early in a sale, the 2010 eighth session, which was substantially earlier in the sale relative to the 2009 eighth session based on the fraction of horses offered, generated much higher sales prices.

The eighth session in 2010 offered yearlings that fell into the range between the first 24.4% of horses offered and those following the first 31.2%.
 
Results from the first six segment-to-segment comparisons (based on percentage completion of each sale) are not surprising.  The two night “super-select” 2010 sessions outdid comparisons to last year.  The four-day select sessions that followed the night sessions started off looking weak, but only because this year’s sessions 3-6 represented similar quality horses, while last year’s sessions started with the higher quality yearlings and proceeded down the quality ladder each day.  Comparisons of 2010 (Sunday) session 7 with yearlings from comparable segments of the 2009 sale showed a slight gain.  Our reported gain of 2.4% compares to the reported gain of “almost 50%” generally reported by other sources for session 7. 

Comparisons of the cumulative averages using our methodology yield a similar story.  Most media outlets reported that the cumulative average was up approximately 18% through Sunday.  Using our approach we calculate an increase of 4.1% through that session.  We have extrapolated what the reports from the media are likely to look like for ensuing days.  Using the session-to-session comparisons reported in most of the media, the reported cumulative averages will continue to overestimate the cumulative average each day, but by a lesser amount each day.  If the favorable numbers we see for the eighth session continue, the final cumulative average will be up, but by single digits, probably closer to 5% than to 10%, and not by the double digit numbers that current media reports might seem to indicate.

Though we would have preferred to have seen double-digit increases, we would view a 5% increase as indicative that the market has bottomed.  We would expect that the November sale, which will sell mares from a significantly smaller supply of mares that were bred in 2010, will show increased average prices over 2009.