I have been developing this system for some time. It trades ES futures on 1 min bars in Tradestation (or Multicharts). The system uses machine learning algorithms to set entry and exit limit prices, which are calculated on the close of each bar. It will typically scalp a tick and hold for 1-2 bars, although sometimes longer. It operates with a $100 stop loss per contract. Results for 1-contract: View attachment 144619 View attachment 144620 View attachment 144621
I should add the following: The system was built using data from 7/2/2008 to 12/31/2012. Results for Jan 2007 to Jun 2008 and Jan 2013 onwards are out of sample. The PNL is net of commissions and exchange fees, calculated at $1.80 per contract. At these volumes it would not be hard to find an FCM willing to charge less than 10c per contract in commission. Alternatively one could rent a CME seat for around $1,800 per month, which would reduce costs even further.
The issue with scalping strategies like this one is: what to do when a trade goes wrong? Scalping strategies that trade with limit prices are looking to scalp the spread (or a little more, from time to time) by buying on the bid or selling on the offer. They rely on markets mean-reverting and most of the time they do. That means that the strategy should make money most of the time - around 90% in this case. But, from time to time, the market takes off in one direction and doesn't look back. The system gets locked into a trade that is hemorrhaging money. At what point do you bail and take a hit? - that decision one of the crucial design factors in a scalping system. As a human trader it is very hard to see a trade on which you were hoping to make $12.50 per contract head South with a loss of $50, $100, or $500 per contract. Sure, the market may come back - but what if it doesn't? That's where I believe machines have an edge - they leave the emotional element out of it and take a view based on the probabilities. The idea is to try to win on average, over a large number of trades, rather than win every trade. Of course, you are going to take a big hit from time to time, but if you can mange those losses you will still come out ahead. You will find that stop-losses always hurt strategy performance. But, like me, you may find them necessary, even so. I know that I would be unable to tolerate a loss of more than $100/contract without wanting to jump in - so I simply set a stop loss at that level, take the hit and move on. That happened a couple of times today. For example, the system was trading well until 2:55, when it got short at 1856.75. The market rallied immediately and the system was forced to bail at 1858.75, for a $100 loss. That doesn't sound like a lot, but it's the equivalent to wiping out around 8 winning trades. Luckily the system trades around 100 times a day so it can recover from these drawdowns, as long as they are not too frequent. It's all about the percentages. View attachment 144624
Someone raised a concern about the low PNL per contract, which is around $6, slightly less than half a tick in the ES. Bear in mind that the PNL is net of commission and exchange fees. It is also worth mentioning that a market making strategy would typically expect to make around $3 per contract. This isn't a market making strategy, since we are only ever on one side of the market. But it does share some similarities, in that we are trading on the offer or the bid. All that said, the concern about the low PNL per contract is reasonable: any deficiencies in the trade execution and the strategy would quickly slip into loss-making territory. that's typically the case for HFT strategies, in general. And that is why, unless they have access to a first class execution platform, the majority of retail investors are better served by sticking to lower frequency strategies for which trade execution, while important, is not mission critical.