The point is to arrive at a number that represents the aggregate performance of all the given metrics (where higher is better).
When you prefer a certain metric to be high (eg %win, gain-loss ratio) then I have multiplied by that number. When you want a certain metric to be low (DD, AV.Loss) then I divide by that number.
The end result doesn’t ‘mean’ anything – it’s just a quick way of aggregating overall performance.
For example, if the risk/reward appears low, it’s not necessarily a bad thing if it’s offset by a high %win.
But, as I said, that formula is only my starting point – I put precisely 3 minutes thought into it.
If you can come up with an alternative, let’s see it! 😁
JSParticipant
Senior
@nonetheless It’s not a criticism just a response to your question whether the formula means anything…
No problem, I didn’t think it was a criticism 😁
The MAR ratio is a good way to compare systems in my opinion
https://www.investopedia.com/terms/m/mar-ratio.asp
MAR = CAGR/DD, so it’s similar to what I’m trying trying to achieve.
I’ve already got the /DD part, but it could be worth adding CAGR to the mix of other factors, perhaps instead of using duration (which is already implied in CAGR) ???
From the PRT report I think CAGR is equal to (Total gains/number of contracts/number of € per point/value of the index) / ((end date of backtest-start date of backtest)/365)
You need to import the report to Excel to do so
For Max DD you also need to normalize it : Max DD norm = abs (Max DD/ number of contracts / number of € per point/value of the index)
For Max DD you also need to normalize it
that’s a good point – I knew there’d be someone who understood statistics better than me 😁
In Excel, the formula I use for compound % is (n2/n1 ^(1/p)) – 1
where n1 is the start value, n2 is the end value and p is the number of compound periods.
I’d assumed that compound annual gain in PRT used something similar, but I’d have to check.
Yes PRT calculates stats in the report too but I prefer doing it myself, because I have not a big confidence in it, for example the sharpe ratios are weird sometimes.
The difficulty is that a system trades with a fixed number of contracts, and the amount of risk taken is different when you buy 1 contract of Nasdaq at 5000 points and 1 contract at 16000 points
Yes PRT calculates stats in the report too but I prefer doing it myself,
Maybe I am not talking about the same, but what I see about these “risk” etc. ratios is that PRT can only calculate them afterwards and applies the math to what happened in practice. This is not really the risk I took – instead it is how it worked out to be (with all being dynamic – just saying).
when you buy 1 contract of Nasdaq at 5000 points and 1 contract at 16000 points
this could be fixed so easily if they made the exposure a constant instead of position size – not terribly difficult. As it is, the Risk/Reward is truly meaningless.
But for present purposes, although the data we get in a backtest report may be ‘flawed’, that’s the data we get. Yes we could export to excel and recalculate everything, but too much hassle for me.
So the problem I’m trying to solve is not how to get good absolute values that might be relevant outside of PRT, but how to arrive at a workable relative value for comparing PRT backtests.
I often look at those numbers and think, ‘the gain/loss ratio is good, but the DD is a bit high’ or ‘risk/reward is poor but the %win is good’ etc etc
What I want is a number that levels that sort of ambivalence.
