[Stocks]

If you were making a model for baseball that involved tracking all 162 games for say Team A. In this model you want game 2 to be worth more then game 1 and game 3 to be worth more then game 2 and so on so there would be more value on recent performance. what's the best way to do that?

I was thinking make every game worth 10% more then the game before like

game 1 = 1
game 2 = (game1)x(10%)
game 3 = (game2)x(10%)
game 4 = (game3)x(10%)

and so on.

Would that work?

[Stocks]

Actually that would probably be

game 1 = 1
game 2 = (game1)x(10%) + game1

so game 2 would be 1x10%=0.10 then add that to game 1 so game 2 equals 1.10

[antonyp22]

Yes that would work. Also have a look at using exponential smoothing.

[Stocks]

Yes that would work. Also have a look at using exponential smoothing.

Awesome that should work perfect man.

The example equation I saw and tried is

F2 = F1 + alpha(A1-F1)

forecast 2 = forecast1 +alpha(actual1-forcast1)

What do you think would be a good value for alpha? The example I saw they used 0.5 for alpha not sure if thats too high for what I would be using it for.

[Baeoz]

What is it you're weighting? Is it the match outcome - win or loss - or something else?

I can think of Linear weighting:

WLin = (tMatch - tStart)/(tEnd - tStart)

Which can be fed into

Log weighting WLog = log(WLin+1)

or

Exponential weighting WExp = exp(WLin)

Of course, if this has nothing to do with what you're talking about, sorry.

[antonyp22]

The alpha depends on how much weighting you want to give to more recent data points. Try a few different values for alpha and see how they would've worked when compared to the actual data values from past seasons.

[antonyp22]

Also consider using "Solver" on Excel to find an optimum value for alpha

[TravisVOX]

Look at exponential decay... you can usually adjust the "half life" to better suit the area you're modeling.