Interesting point that I’ve thought about a lot. If my model is way off, I can almost always find some piece of information that is causing that (injury, etc). There’s always value to knowing limitations of your model and information you aren’t quantifying. I don’t think anyone is doing true Kelly staking; I think a lot of people do some sort of pseudo-fractional Kelly staking where they bet more as their edge increases up to a certain point but then cap their bet size. As long as you do that, even if you don’t recognize the instances your model is missing something major, you’re not going to get THAT hurt by it since you’re not risking your whole bankroll. I’ll add that you can look for these sorts of nonlinearities in the relationship between expected and actual return in data you’re backtesting on.
Interesting point that I’ve thought about a lot. If my model is way off, I can almost always find some piece of information that is causing that (injury, etc). There’s always value to knowing limitations of your model and information you aren’t quantifying. I don’t think anyone is doing true Kelly staking; I think a lot of people do some sort of pseudo-fractional Kelly staking where they bet more as their edge increases up to a certain point but then cap their bet size. As long as you do that, even if you don’t recognize the instances your model is missing something major, you’re not going to get THAT hurt by it since you’re not risking your whole bankroll. I’ll add that you can look for these sorts of nonlinearities in the relationship between expected and actual return in data you’re backtesting on.
@RufusPeabody Hey Rufus, what do you think about @EdMillerPoker essentially saying back testing is a waste?
@RufusPeabody >I don’t think anyone is doing true Kelly staking; I think a lot of people do some sort of pseudo-fractional Kelly staking where they bet more as their edge increases up to a certain point but then cap their bet size. yeh this
@RufusPeabody if your model cant be tracked by seth burn in a google sheet and win against close you cant sit with us
@RufusPeabody think a req. to employ these scaled strats is that the modeler be adequately backtesting, otherwise just button-clicking imo can imagine scenarios when weighting could yield significant Δ to market well outside of discretionary sizing caps but still not enough to deter betting
@RufusPeabody how do you backtest load management? ( i only know excel.exe)
@RufusPeabody based on your experience/observation, would you accept or reject the hypothesis in quotes here? and yes, defining "too large" is kinda necessary for this to mean anything lol
@RufusPeabody based on your experience/observation, would you accept or reject the hypothesis in quotes here? and yes, defining "too large" is kinda necessary for this to mean anything lol
@RufusPeabody This happened to me in the NFL last season, where you stake more based on a greater edge. Turned out I had a worse winning percentage, on those bets for the first time in a long, long time. It got me thinking about "yoshing" this upcoming season.