Chance of losing in a month, year, etc.

[Rick]

The first two months of this year have been pretty disappointing for me, only breaking even for 195 plays. So, I wondered how likely that really is. I calculated the standard deviation of the previous year's payoffs using a spreadsheet as the first step. I took that and divided by two to get the standard deviation of $1 payoffs. That turned out to be 2.9, and I'll call that S. Also, last year I had 25.1% winners and 13.65% profit on 1203 plays, so I know that the expected value of a $1 play is +.1365, and I'll call that E.

Next, I calculate the standard error of the mean for the $1 payoffs in N plays, which is S/Sqrt(N). For 195 plays that turns out to be 2.9/Sqrt(195) = .65. Looking up the one tail probability of that under the Normal Distribution gives about a 26% chance of that occurring. So, it's not really all that unusual. For an entire year of 1203 plays it comes out at about a 5.2% chance, and for a month of 100 plays it comes out at 32%.

Now everyone's data will be different, but I think that it's obvious that if you're going to play this game for long, you'd better be ready to go for quite a while without winning anything once in a while. One losing year out of twenty seems like a reasonable risk to take but I certainly wouldn't depend on it for a monthly income with about a one in three chance of not making anything at all in a given month.

You can also use the above information to figure a "true" Kelly percentage. I say "true" because the often quoted method using average odds and win % is incorrect in assuming a constant payout. The true Kelly is computed as E divided by the Variance, which we can get by squaring the standard deviation. So the Variance, V = 8.41. Then, E/V = .1365/8.41 = .0162. The incorrect Kelly proportion, assuming constant payoff, comes out at .0386. So, if you calculate it incorrectly, you'll be overbetting by more than twice and that's enough to turn a winning method into a loser.

I hope someone finds this all interesting.

[sjk]

I calculated the standard deviation of my returns for last year (normalized to $1 bets) and got 5.13. I have never won every month in a calendar year (maybe this year) and as your post implies, a losing month is not necessarily an indication that mistakes were made.

That is one reason I wondered about the effectiveness of the AI approach in another thread. If my understanding of the AI approach is correct, I would evaluate today's results to learn what went right and what went wrong and make adjustments to how I handicap tomorrow. With the variability inherent to the process I could have an 0 for 30 day and I would not be able to draw the conclusion that I had done anything wrong.

If you base a program on what you learn from 250,000 races in the past, what you learn from the 140 of so races to be run today would not seem to add much to your knowledge of what goes on in a race.

[Rick]

Exactly right. One of the hardest things for me to learn was not to fiddle with my method after seeing a couple of hundred races with subpar performance. While I think all methods degrade over time to some degree, it usually takes many years before a good approach becomes ineffective. The best example of this is what happened with early pace methods from the 80's to now. The value of that approach is much less now than it was then but it did take a remarkably long time for the public to catch on.

[Tom]

I have no hard stats to back it up, but I know enough to wind down after the Breeder's Cup and not get to into it until Mach-April. I generally do well at Florida tracks in the winter, but the thing is I don't play at all ubless I happen to be having a very comfortable day and "feel" like playing. I usually don't in the winter. My play drops to a few races a day at best because I do not usually feel like doing a lot of work aroudn the holidays and January-February is usually my busiest time of the year at work. When the days get warmer, I start firing up the brain for some serious play, but history tells me that winter brings far more losing plays than I care to male. I just don't have the focus to play hard more than a day or two in a row. Beside, the constant cancellations, off tracks, yadda yadda yadda just make my grumpy. Right now, if I turn on TVG or HRTV and see an interseting race, I will look at it and play it if I have time, but no more "getting ready" for a day of racing.
Less play annually hase improved by bottom line.

[Rick]

Update: I'm now up 13.46% for the year on overlays, so the first couple of months does seem to be just an unlucky streak. Not as good as last year yet but in the same general range, especially considering the large variances in horse racing results.

Also, an idea I mentioned on another post of using lower weight this race compared to last proved not to be helpful over a larger sample. It wound up with 28.67% wins and 2.87% loss compared with 30.56% wins and 5.03% profit playing every race with the original method. Another interesting study I did involved top BRIS Prime Power horses that were also my method's top selection. That wound up with 39.72% wins and 0.98% loss, so that would not be helpful to the ROI either. The much higher win % was somewhat unexpected though, since I assumed that BRIS puts everything but the kitchen sink into that rating and I'm only using BRIS data myself. It appears that they may not be using some of their own most useful data.

[kenwoodallpromos]

I'm no math whiz, but I understand that!
I calculate each race as to its level of predictability, which could change depending on the weather, other scratches, race placement, speed-favoring, etc. Even using exactly the right bet using the right factors and angles, the predictability could change. And the higher quality the competition, the less has to change to make the race more unpredictable

[JPinMaryland]

If you base a program on what you learn from 250,000 races in the past, what you learn from the 140 of so races to be run today would not seem to add much to your knowledge of what goes on in a race.


this would only be true if the medium your are studying is not dynamic. I suggest that horse racing, like any sport with mortal beings, is going to be dynamic.

So more timely information could be more highly relevant to today's race. It is easy to think of an example such as a race run over the same track earlier in the card, but there could be other more subtle issues. Such as changes in breeding patterns; changes in average number in starting field; change in drug policy, chagnes in equipment, training methods, track surface, etc.

ANy change in any of these could produce patterns whose recent trends may be more relevant than distant past ones.

[sjk]

JP,

I don't find today's races any different from those of six or seven years ago; at any rate I am using the same program (with a few incremental improvements along the way) with similar or better results.

Here's a position which even more people will probably disagree with: I don't find there's much difference between cheap horses and stakes horses; when I bet a race I am generally ignorant of the type of race I am betting and the distiction my program makes is fairly subtle and probably doesn't especially matter. It makes no distinction at all between races for young horses and those for older horses.

It has often crossed my mind to try to use info from earlier in a day's card to improve my results but it has always seemed impractical to do so. If I spent time collecting data as the races take place, it would inevitably make it more difficult to play as many tracks as I can by ignoring that information.

[Rick]

sjk,

You're totally right. There haven't been any significant changes in the last 6-7 years in horse racing. Until there is some new information available, that's not likely to change. Yeah, horse racing is different now than it was 15-20 years ago, but what really has changed lately or is likely to change in the near future?

I do think that some tracks are a little more predictable than others if they have longer meets and that fact may offer some advantage in predictability. But trying to predict what will happen today or tomorrow from a pace bias is not very productive in my opinion. In fact, I think these days that track management actively tries to eliminate these biases whenever possible. Unfortunately, most tracks are pretty much the same these days and it's difficult to find any kind of bias that remains undetected for long.

The big secret that none of the famous authors wants you to know is that the methods that worked in the 80's don't work any more. Yes, they made a lot of easy money back then on identifying biases, but they don't make anything on betting any more. Now they're doing books or seminars or real estate and still trying to convince you that they know how to win. Well, they don't because they haven't kept up with the game.