Time Decay Model

[Dave Schwartz]

I want to build a "time-decay" model.

Can any of you suggest a website that might explain some different schemes for this?

Dave

[46zilzal]

I want to build a "time-decay" model.

Can any of you suggest a website that might explain some different schemes for this?

Dave
sounds like me teeth and my vision!!

[Dave Schwartz]

groan <G>

[highnote]

I want to build a "time-decay" model.

Can any of you suggest a website that might explain some different schemes for this?

Dave


I don't know any websites, but it's a very interesting notion. I'm giving some thought to how I'd do it. If you're interested, I'll be glad to post it. I don't have time to do the work of researching a time-decay model myself (I don't know what the hell I'd use it for), but I would be interested in the value you come up with. Sounds like an interesting challenge.

John

[mcikey01]

TRY THIS, DAVE: http://daytrading.about.com/cs/options/a/black_scholes.htm

[mcikey01]

Dave: I should have asked , "Time decay in relation to what aspect of handicapping or wagering?"

[Dave Schwartz]

Mickey,

I should have asked , "Time decay in relation to what aspect of handicapping or wagering?"

I could tell you but then I would have to have you killed. <G>

Seriously, track models are where my interest is (for now).


And thanks for the link. The story behind Black-Scholes is an interesting one.

Dave

[Handiman]

Sure you didn't mean Tooth Decay? :lol:


I can't seem to understand what relationship time decay, as it's normally associated with Options, would have to do with horseracing. The only possible relationship I can possibly see, would be the effect of days away as a cause of decay as it relates to a horse losing sharpness.

Just thinking out loud. I do like to see people exploring new avenues of development in this crazy sport. I call it crazy, cause it's the sport of Kings, and most royal beings were the result on serious blue blood inbreading. But that's for another topic.

Handi

[highnote]

I think what David is getting at is that each race back in a horse's past performance has some significance. The more recent a race the more significant. Question is, how significant is the fourth or fifth race back?

Logically, an older race has some significance, but probably not as much as a more recent race -- all else being equal.

When you develop a handicapping model you throw all kinds of stuff at the wall. Then, you keep whatever sticks.

I have no doubt it will help with the model, but as my aunt used to say, "Sounds like a lot of shit, mess and bother." :D I love that line.

[toetoe]

Formwise, recency is everything. However, when a win drops off the pp's, being more than ten races ago, it can hide a non-form-related pattern match, the result being a nice longshot. I'm a big formophile, but I don't see how we can make a sliding form scale. The form is acceptable or unacceptable, and we move on to the other factors, or we throw the horse out.

[linrom1]

Sure you didn't mean Tooth Decay? :lol:


I can't seem to understand what relationship time decay, as it's normally associated with Options, would have to do with horseracing. The only possible relationship I can possibly see, would be the effect of days away as a cause of decay as it relates to a horse losing sharpness.

Just thinking out loud. I do like to see people exploring new avenues of development in this crazy sport. I call it crazy, cause it's the sport of Kings, and most royal beings were the result on serious blue blood inbreading. But that's for another topic.

Handi

An interesting principle behind time decay as it relates to options is the concept behind Maximum Pain. This is the point where most options players lose regardless if they are long or short the option. :bang:

[garyoz]

Two comments:

If you are taking a non-linear regression approach (ala Black and Scholes) you are going to run into some problems of correlation in the predictor variables if you are building it from the ground up.

Also one of the problems in applying time series analysis approaches to horse racing is that you will probably do some type of "smoothing" that will knock out subtle variance in the measures. They may be the forward moves or other peak performances that are indicative of a winning effort. This is why moving averages (IMHO) just muddle the data in racing and probably just move the measurement toward the par of the class level.

If you want to use some type of decay approach, you may just try to identify a curve that fits the data and not worry about tying it to explanatory variables. They are a bunch of canned software programs that provide curve fitting. This link popped up early in a google search which uses GraphPad which is a pretty good program. http://www.curvefit.com/how_to_derive.htm

[Dave Schwartz]

Okay, I'll give you guys this much... The time decay is for building a current track profile.


Dave

[rrbauer]

Aha!

A function of weather and track maintenance procedures.

[highnote]

If you want to use some type of decay approach, you may just try to identify a curve that fits the data and not worry about tying it to explanatory variables. They are a bunch of canned software programs that provide curve fitting. This link popped up early in a google search which uses GraphPad which is a pretty good program. http://www.curvefit.com/how_to_derive.htm

Dave,
This is pretty much what I came up with... plot the data and then find an equation that fits it.

You might want to checkout "Handbook of Nonlinear Regression Models" by Ratkowsky for a good explanation on how to fit an equation to a curve.

John