Runs of “up” days in the S&P, from Big Al
An analysis of runs of "up" days (i.e., Close-Close change is positive) in the S&P, through 2 May of this year:
And then I asked an AI to model flipping a coin biased in the same way (53.7% heads), and you can see the results here.
Anatoly Veltman comments:
Yes you're putting numbers out - no complaint there. Huge complaint on the premise: why would 9-day "run" into tomorrow bear same fruit as some totally different 9-day "run"?? What's a "run"; why would different-size price increases under all different relevant variables have the same impact on further trajectory - just because you assigned the same "9-day length" value to current "run"??? IMHO this sort of input can't be expected to help much.
Big Al responds:
That's actually the point: when we assign importance to runs of days, we have to be careful because the run distribution in actual data looks like the distribution you get doing a coin-tossing exercise with a market-biased coin. Which doesn't mean that analyzing runs can't produce anything useful, just that there is a high hurdle.
Anatoly Veltman adds:
My point is mostly about DIFFERENTLY-SIZED up-days. Some days could've been up $2, while others $200…Some days might have been not up-days in SP500 index, while up-days in SP500 futures. And dozen other variabilities that would make one "9-day run" be vastly different from other "9-day run" in impact on future expectations. Not that there are many ideal ways of Input, but this sort may just be prohibitively flawed.