ZTR Money Management and Algo Optimization …
- This topic has 16 replies, 5 voices, and was last updated 7 years ago by
.
-
I think that to judge if the Algo is good or not, we need some Criterias
For me a Good Algo must have
1 / >70 % Winning trades
2/ < 25 % Max drawdown
3 / VQN >2
If your have some others criterias ….
1 user thanked author for this post.
Hi – Thanks for starting this thread. Hopefully it gives a nice guide for people to refer to once we get some consensus around what is the perceived “standard”.
Re the judgement of results – Yes, I agree on the 25% max draw. I am a bit flexible on the winning % as I think it can be between 50-75% as a high gain/loss ratio offsets that to a certain degree, but would depend upon the overall results and strategy. That said, the VQN number and a minimum hurdle score for that like you have stated of 2 should address that to a large extent. Other parameters I would personally also look for are :
1)Maximum consecutive losses – ideally around 3-7. Since if you are risking on average 1-3% per trade you are likely to incur a draw of upto 20% worst case which still leaves you breathing room for a recovery based on your backtesting without wiping you out substantially.
2)Loss of worst trade – should be reasonable and less than the profit of the best trade. Shows if you are taking on more risk than you originally thought. Hard to put a number on it as it depends upon strategy but again no more than 5% seems reasonable.
3) Average time in mkt – Is a mark of how efficient your strategy is when entering and exiting and the accuracy of its timing. Also warns of hidden cost of capital costs eg. if it takes 2 weeks to make 0.2%, your profits will be eroded rapidly by financing costs. You want to be ideally around 10% but below 20% ie equivalent of being in the market continuously for 1 full day out of every 5 trading days.
4) Distribution of profits and losses – Unfortunately PRT at the moment does not provide a histogram of these as it would be useful to see if your results have a few large trades on either side that skew the overall number. A good strategy should have a fairly even normal distribution of results which would increase confidence in the overall system. Until we have a easy way of doing this (we can do it manually in excel) we have to “eyeball” the Abs Perf column in the backtest to make sure there is nothing weird going on. You can also look at the daily results graph chart for a quick evaluation but of course that may not be accurate if you have multiple trades on the same day so it’s not ideal.
5) Consistency of results – I would want to see the Quarterly results histogram on the backtest and look for a fairly even performance without too many large peaks and troughs. Monthly/weekly chart has too much noise and Yearly masks issues if there any. You would expect some bad periods as every system has them but overall (esp if you have a long term system) you should place a higher value on low volatility than you do on bigger profits. This would also suggest that the system is robust enough to cope in most types of market.
Please note that it is not my belief that every system has to comply with all the “nice to haves” above. There are perfectly good systems that don’t necessarily fit all of the above but work just as well, it’s a overall judgement call based on a lot of factors. Other thoughts to follow…..
Hi Grahal,
Max drawdown, is max drawdown (You have many informations on the net). You have it on the results after on optimization. This is the maximum loss/drawdown during the optimization. If yur capital+profit=10000 euros. With a Max DD you can’t lose more than 2500 euros
For the VQN (I was drunk …), sorry it’s SQN of Van Tharp
http://www.vantharp.com/tharp-concepts.asp#SQN
It’s a very interesting concept who estimate the distribution of the Gain. I find it’s a most have indicator
https://trading-journal-spreadsheet.com/faqs/sqn-created-by-van-k-tharp/
2 is the strict minimum to estimate a good/low distribution of price
It’s so useful I hope PRT we had it on the panel of results in a future version
Van Tharp is a most have to read for all Algo optimizer
If you have a Max DD<25 %, SQN>2, a winning strategy …You can have an intersting strategy
For The Walk Forward optimization we continue to work on it, but I have some ideas I need to confirm
See U
1 user thanked author for this post.
FYI, and I already talked about it in some other threads, PRT is currently working on a whole new data engine for faster executions. I had a talk this week with a dev who told me that CALL function should be thousands times faster than now. It opens great potential because we’ll could use CALL to make external re-usable functions instead of copy/paste many times the same code for instance. Don’t ask me about ETA, I dunno.
Find exclusive trading pro-tools on 
