If anyone is interested, I coded up the simple system described at the start of this thread on MT4. The reference looked reasonably professional, the claimed results were good, though on first look the system looked junk. If Opening 5 min candle's up then Buy 2nd candle's open + SL @ 1st Candle's Low and TP @ 10R If Opening 5 min candle's down then Sell 2nd candle's open + SL @ 1st Candle's High and TP @ 10R Else do Nothing. As far as I can see, it ran fine on MT4, and I tested it from 2nd Jan 2023 to Friday 14th April 2023. Assumed spread was 1.2 on the NASDAQ. The results; in 73 trades, a total loss of -267.1 points, an average loss of -3.66 points (3 times the spread). Maximum run of losses 18 (3 and a half weeks of solid losers!). All in all one of the most lousy systems I have looked at. I might have done something wrong, but I doubt it. Intuitively the system looked junk, and so it proved. Maybe it had a bad patch; maybe the results for the previous few years were as good as the report says. It doesn't seem likely to me. If someone else wants to have a go, give it a try. It's an easy enough thing to code if you have a test bed of some sort already in MT4 or similar. That old adage that you should test things yourself before believing them seems to hold good here.
I also tested it with ES, from October 1rst 2020 to December 30th 2022 (27 months). Using the exact same parameters as defined in this paper and using reliable tick data. Results: Average of 45 pts/month Max DD: 291 pts (from 2022-04-25 to 2022-05-12), time to recover: 2.5 months Best month: 353 pts (Oct 2022), 218 pts (Dec 2021) Worst month: -177 pts (Nov 2022), -137 pts (Mar 2021) % of profitable month: 78% Conclusion: It's not that bad, although the DD is way too much with a recovery average period of 6.5 months. It's a simple and effective algo but the management of loss should be improved. I'd say it's a good start for someone looking for an profitable algo. I haven't tested it on NQ as in the paper (cause I don't have the data), but I guess the results would be similar.
Thanks for this. Yes the equity curve looks too bumpy to me, even if there is positive expectancy which I still doubt. The data you used was inside the original test data. From experience it's when you forward test on new data that the results get more significant.
Yes, you are right about the test data. I didn't pay attention to the period that was used in the paper. I just retested it using Jan 1rst 2023 to April 17th 2023 data and here are the results: Total: -129 pts Jan: -118 pts Feb: -110 pts Mar: 89 pts Apr: 10 pts Max DD: 258 pts It's not getting better...
I searched through that paper, but they don't seem to use a factor for spread and associated costs. That could (unfortunately) explain the results for small 'stoplosses'. A good thing though - at first glance - seems to be that there is low correlation with the index itself: