How does skewness in data affect the backtest result?

Algos, strategies, code
1

Almost all stocks have either a positive skew or negative skew. My question is how will this affect the results of my backtest compared to if the skew did not exist?

2

What’s the slippage in the skew that you are talking about?

1 Like
3

  1. Any edge should work better than the baseline/buy and hold. If market already gives on average say 6% over a period then edge should do better. This may not be as black and white as we also have DD to consider.
    For active trading, we can look at the average per bar returns of the market and any possible edge has to do decently better than that over the holding period. So for ex, if you buy on dip, you can look at say 5/10/20 bars post entry and check what returns you get vs 5/10/20 * per bar baseline returns.

  2. If market has a positive bias, then it might be harder to get edges to work on short side. Even if it does better than baseline, you might still lose by losing less. So absolute returns also matter, obviously.

1 Like
4

What is a slippage in the skew? Are you referring to slippage in terms of impact cost?

To rephrase, I want to know, how would my backtest result be different if the data were positively/negatively skewed vs if it was normally distributed? Why does knowing skewness in the data matter and how does it help?

5

Thanks, one question that strikes me is what is the need to measure Drawdown if you have a long strategy where you set a stop that is say within 5% below your buy price and you ignore trades that have a much wider stop, how does drawdown matter here?

6

Your stop can get hit say 5 times in a row or some other mix of wins and losses. Stop is not magic, there is a tradeoff. Anyway, you need to measure risk in some way and dd seems the most obvious way for trading. How much i might lose from equity top during bad phases? Everyone has to go through dd ( except liars or perhaps HFT/arbitrage). Only way to mitigate it a bit is to trade multiple systems/timeframes/markets.

Although its easy to over optimize rules around drawdowns, esp if sample size is low or if there is one really bad outlier. This can hide the risk but market ofc does not care. One decent thumbrule is to allow twice the tested DD along with some common sense when looking at the data.

1 Like