I'm playing around with a new system. It's based on a pairs trading trend following strategy using a single highly optimized input parameter. The optimization/backtesting was from 06-12 with 13-15 being the out of sample data. It has a very limited set of pairs that it works on (about 40 pairs) but gives at least a couple signals a day. The average trade duration is 2.5 hrs (4 hrs for the longest trade), the slippage/commission used in the simulation 40 bucks since there's four legs to each trade. It only trades liquid symbols (avg daily volume > 3 mill shares). It has a fixed position size (5k per symbol levered 1:5 via portfolio margin, so 50k total position size, ~10k buying power reduction), no beta weighting on the individual stocks but dollar neutralized (so if one symbol trades at 50 and the other one at 100 it'll trade twice as many shares of the 50$ one). All of the pairs have > 1100 trades (horizontal) with commission and slippage eating away most of the real profits. The trades from ~900/950 to 1100 are the walk forward trades. Here's the challenge, since the equity curves look good (it is not a look ahead issue on some but not all pair candidates, the question is if it's too much optimized? I guess the real question is, should a good system work on all underlyings or can you hand pick some where it works and just trade those? Here are a couple of examples
The very 1st question that comes to me is : - what is the average net per trade (#ticks per share) I am far from a specialist of stock trading, so excuse me if the next question is dumb: - since it is pair trading, every single trade has a short ... is shares-availability for these short a given? How reliable is your backtest? If I remember correctly, LMT orders have no guarantee to get filled for stocks, even when historical data shows price going through the LMT (unless you are clearly on the wrong side of the trade) What is the aggregate P&L curve for the 40 pairs? What is the largest combined position across the 40 pairs? If that exceeds your capital, how would you select between concurrent opportunities?
If your results produce those graphs -- then yes. -- life is pretty simple, not everything has to be complicated.
Good questions, for many I actually don't have an answer. I haven't calculated the average ticks per share, I guess it highly depends on the initial cost basis (it'll be different for a 20$ stock than for a 200$ stock). The total average $ per trade is 48$ the median over ~ 49000 trades is 47.8 (after fees). You're right every trade has a short trade and a long trade. The short availability is pretty much a given. I've been pairs trading for many years and rarely run into situations where there's no shorts available (I trade via IB and since the size if very small (25k per leg) I won't make a dent in their short inventory). If something goes bad (e.g. there's no stock to short my existing pairs trading logic alerts me. All the symbols should have shorts (IB publishes a list), however sometimes it did happen in the past so it may be a case where I've to close the trade very quickly. My other pairs trading strategy holds much longer and it's more likely that I need to buy back a short by the end of the day cause of whatever reason than there not being a short available at least based on the universe of stocks I'm using. The tricky question is always how reliable the backtests are. In this case it seems pretty straightforward, also the commission/slippage is somewhat conservative. My other pairs trading system basically sits on the bid/ask and tries to execute the second order once the first one gets filled, in reality that should be much less than the 40$ guesstimate (this is like a worse case assumption, in reality I deal with ~ 10-15$ slippages/fees when I trade those sizes). That also accounts for odd lot fills. During the test the maximum nbr. of signals I've received was on 12 pairs and in real life that would not cause any heart burn, it would increase the leverage of my account but even 24 concurrent signals would not make much difference or cause a margin call especially since the holding period has a time stop of 4 hrs. The real question is just whether or not the system is viable if it just works on a subset of the symbols but I guess I'll have to turn it on to get that data point (which I will regardless) unless someone has a good justification as to why a robust system should/would need to work on a broader set of symbols.
Watch out for entries on spikes, run a comparison against average bar price to detect unrealistic spikes. Entries on spikes are the worst enemy for equity systems (all those dip buying strategies that don't work in real life). If you cherry picked the pairs then it's a clear cut curve fit, unless you're dealing with very broad indexes. Alternatively you could pick pairs that theoretically should work from another exchange, then run it and if the results are equally as good you're good to go. Make sure you don't look at the performance before running the strat.
Some of the pairs are indexes, some of them are sector related and some of them are not related at all. I eliminated the unrelated ones, e.g. AA and BAC since there's no reason why they would work even though the equity curve looks good. I've keep the sector related ones cause for those I can at least come up with a reason why they would work. I guess I won't know if they work until I try them, so I'll just reduce the position size by half and let them run for a month (for 50 or 60 trades).
It looks like an over-fitted A.I. System, but if its not I would trade it if the average Trade covers more than the cost of trading (slippage, etc.).
While those equity curves look to be profitable, it looks like a pretty slow walk higher. Watch out for systems with a low payoff ratio and a high percentage of winners. If the system degrades even slightly, you will go sideways or down in short order. I should know as I have trade a lot of those type of systems. I am now splitting up my trading with a few breakout systems. The breakout systems will give me a chance to get ahead with some big winners.