Good morning everyone, this is a copy of a conversation published on the Italian section, copied here because some users who speak only english, were interested to join and paticipate.
It will be my care, to report interesting solutions or tricks that will be proposed here on the Italian section and viceversa.
I’ve noticed for a while now that the backtesting in PRT (especially if the algo has many configuration parameters) is less reliable than in other platforms. Not because it makes mistakes, but because it discards a lot of results, and among those discarded there is often the most interesting one in terms of performance/risk ratio.
So, I started thinking about reducing the backtesting period. The problem, however, is that it is universally recognized that the longer the test period, the more the chosen parameters will be able to provide better average results. But this clashes with what I mentioned above: if PRT discards more than 50% of useful results, I risk losing the most balanced parameter combination.
At this point, I thought I could develop a method that identifies, in the past, the periods most similar to the current one and then use those periods as backtests.
But how can I select the periods most similar to the present?
I’m trying to apply this method:
I built an indicator based on ATR that tells me the monthly volatility. Then I measure the volatility highs and lows every six months and choose for the backtest the period with a range most similar to the current one. Each month I repeat the measurement of the last six months and redo the backtest looking for a similar period.
What do you think?
Here comes the question: what other parameter besides volatility could I include in the indicator to identify periods in which a stock’s behavior was similar to that of the last six months?
Also, do you think six months is the right interval? Why not three or one?
Thanks to anyone who replies.
Best regards
.
.
