Hi trignal67, and thanks for your constructive questioning. This algo was formed from an 'idea first' viewpoint observing the market. I'm usually a 'data first' guy, and all of my FX algos are derived from a 'data first' approach. Both these terms are very well presented by Robert Carver in his 'Systematic Trading' book, which is a great read. So in this case, the idea was formed from looking at how price reacts around the blue line you see in the chart. It's an average of my own design, based on a Holts average, with some modifications. My observation was that in many cases, price reacted in a certain way around that line when it deviated by some amount of ticks, i.e. reverted to that mean. Taking that observation, I coded the algo and tested it on a 5000 bar historical dataset in order to ascertain the reversion stats. Around 4000 bars were dedicated to back test, and after some small modifications, a forward test on the out of sample 1000 bars. Now, there will be those that feel back testing needs way more data points, there are those that will say back testing is worthless. Everyone has their view. In my case, I need enough data, to see something that is statistically relevant and gives me a warm fuzzy feeling inside. I'm not a statistician by any means, but I know enough to know what I know works for me. The testing results were positive. And good enough to make me put a grand into this adventure. In the scheme of things, it's not a lot of money. But, it's my money, I earned it and saved it, and so to me it is a serious endeavour. I'm not going to divulge how profitable the testing was, because that is really meaningless. Going forward, the market will give what's it's going to give, I cannot force it to perform as it did historically. I think that's a really important point for me anyway, that my back test only tells me that the strategy has potential. The forward test only serves to add some confirmation to that potential, enough to bring me to a decision of whether to trade it or not. The algo performs well over that sample dataset with 3 different Risk to Reward profiles: 1.5:1, 1:1 and 1:1.5. These are the three profiles I generally stick to. I like the 1.5:1 risk:reward, as it results in nice runs of wins, and that brings me psychological comfort. Over the coming weeks, I may run all 3 variations on this one account, and although entries will all be correlated, exits will be per the reward ratio. But that will need an additional deposit, or the risk across a single set of entries will be around 9%, which is even too much for me. The rationale for the journal is simple. As a hobby trader, I'd like to expand my horizons a little (e.g. this is the first time I've traded futures) and engage in a some constructive discussion around the stuff I'm doing. That may not provide all the detail you've asked about, but I hope it adds some context.
I'm just adding some further detail with respect to my post above, as it's a little unfair of me to quote risk:reward ratios without also quoting the win rate expected from the tests at those ratios. With the current risk to reward of 1.5:1, the win rate over the back and forward tested sample period is 67%. That's 7% better than the implied odds. With the risk to reward set to 1:1, the win rate over the back and forward tested sample period is 56%. That's 6% better than the implied odds. With the risk to reward set to 1:1.5, the win rate over the back and forward tested sample period is 46.5%. That's 6.5% better than the implied odds. So the million dollar question is whether the win rate will hold up going forward. And therein lies the rub. But I guess that's why we trade, to challenge our view of the market; to attempt to beat it, to win. But she's a fickle mistress.
The market giveth, and the market taketh away. Today the win rate swayed to 50%, three losers in a row also seen here, followed by 3 winners, but not enough to save the day. Running PL -$18.07
A tough gig today. The astute will notice that this v1.1 of the algo, with 2 minor changes. The start time has changed to 1400GMT and the I've switched the Risk:Reward ratio to a positive 1:1.5 after reviewing some of the max favourable excursion (MFE) stats from the back test where I feel I'm leaving some money on the table. Today though, has not been favourable! Running PL: -$85.75 (within drawdown expectations)
I can confirm this @MarkBrown as I also use (on occasion) MT5 with AMP. It's real. Here's what mine looks like.
Thanks for the info. It's interesting to think about how reversion to mean algorithms could work. The main problem is how to recognize a one-way trending market and step aside and go with the trend. Since markets are mean reverting most of the time (appx. 70% by some estimates) It seems like we need something that will tell us when to stop using this strategy. If prices have gotten to an extreme from your blue line, how can we tell our algorithm to stop trading? Two losses in a row? The worst drawdowns in this system will be on a day when we keep going against a strong trend. Ideas?
Can we come up with something on a shorter timeframe that tells us that we are about to revert to a mean? The blue line. Also once we have our reversion can we have a fail safe that will get us out of our position. Seems like a fixed stop and target are inadequate for this purpose. Also, can we look inside the 5min bars to see if a reversion is about to take place rather than rely on a fixed excursion from the blue line?
Hi trignal67, Absolutely valid comments and suggestions. I actually started looking at something very similar yesterday after a couple of fairly strong trend days on the 5m bars. I have so far evaluated looking at price action to ascertain if we should take the mean reversion trade ,e.g. a bar in the opposite direction, and have looked at a shorter timeframe indicator to filter the reversion trade based on the trending bars. Neither of those provided any higher level of profitability on the back test, but the price action filter did reduce the draw down, at the same time reducing the number of trades taken, while still showing a good return over the sample. So I am examining this aspect further. I can of course 'just go with it' and accept the losses when they occur, based on the test stats. That was partly why I changed the RR to the positive ratio, due the MFE stats offering more to be had from each trade, further offsetting the losers. Thanks for your comments.
End of the week, only 2 trades today, both losers. There was one trade triggered just before the open due to a small bug in the algo, now resolved. it was positive, but doesn't really count. Running PL -$135.47 (within model draw dawn limits) Next week, I'll be running v1.2 of the algo, which has a price action filter applied to the mean reversion opportunities to help with strong trending 5m bar deviations from the mean. I've also allocated another $1000 to this algo, reducing the risk from 3% to 1.5% per trade. It's sitting in the brokerage account, and will only be allocated to the trading account if required. Have a great weekend.