positive/negative days

[TaranveerSingh]

Trying out different ways to punish volatility while ranking stocks in my rotational systems.
Could someone be kind enough to share a code for counting positive and negative days in a given lookback?

[GlenPeake]

Have a read of this thread Taranveer, Julian posted some code at the end of the thread which might be what you’re looking for.

https://edu.thechartist.com.au/kunena/trading-system-idea-s/525-more-rotational-ideas.html#7582

[JulianCohen]

Honestly I couldn’t remember that! Even when I looked at my own post…..

[TaranveerSingh]

Thanks guys.
Tried many variants. While rewarding volatility destroys performance (dividing the ranking metric by positivecount/negativecount), punishing volatility doesn’t do much.

Something I didn’t expect to conclude –
There’s ROC/ATR ranking used in the forum. I thought this ATR represented volatility. But when you use ATR as a percent of close, it downgrades the performance. I feel what works in ROC /ATR isn’t the volatility part but the absolute price bit (it rewards lower priced stocks). Next I tested for ROC/C and that did increase performance compared to ROC ranking.

Naturally, tried reducing max price in the price filter but somehow it didn’t do much.

Out of curiosity, I also tested how my strategy fares if I reversed my filters (tried say CMA) to see if they mattered in the first place. Glad to say all of the filters are relevant.

[ScottMcNab]

Are you using adjusted or unadjusted close for the ranking Taranveer? There is a risk of a look ahead error. This also impacts atr so it would also need to be adjusted I believe. In practical terms, stocks that are going to be successful in the future and run up in price ( and undergo split(s) ) would be selected preferentially (and incorrectly) in the backtest if ranking ROC/atr or ROC/C were used….. giving a backtest with potentially unrealistic results (as the denominator would be made smaller when adjusted for the split I think).

so I dont think that atr as a percentage of close downgrades performance as such but rather it removes the look ahead error ?

[TaranveerSingh]

Scott, it just “happens” that I use unadjusted but I didn’t think of what you pointed out. Thanks as always.

[Author]

Posts