Was curious at what kinds of very high (above 6, let's say) profit factors people have found backtesting systems of over 100 trades that actually turned out not to work in real-time. I'm thinking that once a system gets over a certain number of trades, it becomes increasingly difficult to sustain a high profit factor unless there is something the system actually captures in the way of a market feature. Seems to me that it would be nearly impossible to "fake" a profit factor that high with any consistent set of rules. I know that 100 trades isn't really all that many, but if you had a system which you knew would have a profit factor of 6, or close to it, for its next 100 trades, I doubt you'd need more than that 100 trades, levered as high as you could, to make yourself as much money as you'd need for a long time.
Neither very high nor very low pf systems are robust. In my experience the most robust systems maintain a pf between 1.8 and 2.4. Those other high pf systems are curve-fitted and optimized. Please understand the math. A system with 100 trades all winning has infinite pf. If the trade 101 is a loser and let us assume R:R =1 then the pf becomes 100. If the trade 102 is also a loser, pf becomes 50. Here how it goes: 100 winner and then add consecutive losers: infinity 100 50 33.33 25 16.6 14.3 12.5 11.1 10 9.1 8.33 7.69 7.14 6.66 6.25 etc. You see that from infinity to pf=10 is only 10 losers way. This should tell you something. This game is really tough.
I would be the last person to say trading isn't tough. But, it's kind of a weird discipline in that it is tough/practically impossible up until the moment you know what you are doing. Then, it merely becomes difficult. Anyway, the system in question is definitely optimized. One half of the set of trades had a profit factor of over 10. That was in an environment that was obviously perfectly suited to the type of trade this system makes. But, even the other half of this set of trades, which has been in a much different market environment, has a profit factor of about 3.5. I even know the date when the market environments switched, so I can segment the trade data by that. So, even if the initial trades used for the optimization were in a "perfect" environment and that environment was unsustainable, the post-optimization profit factor is still 3.5. What I would ultimately like to know is if this 3.5 is sustainable as a "minimum" future profit factor. Also, given that the first environment was one of rising volatility and the second environment was one of falling volatility, does that mean that the system has experienced a complete market cycle and that the future profit factor will sometimes be ~10 (when the market is volatile) and sometimes be ~3.5 (when the market is less volatile)? The other type of market environment, which we haven't experienced, would be a range-bound market, but since this is not a breakout system, I can't see how that would hurt it to the extent that the profit factor dropped below 1, especially since I could probably do further segmentation on the data and find miniature "ranging" environments and test what has happened during them. It's doubtful that the profit factor for any subset of these 100 trades has been under 1 for very long. Obviously, there's no way to know the future and the market could come up with some environment this system can't handle. I was just curious if there isn't some point at which a very high profit factor ceases to be a reflection of obvious curve-fitting and becomes a reflection of a robust system.
I have a funny example relating to PF. I have a system when tested on different indexes, like S&P or nasdaq or other, has profit factor ranging from 2.5 to 4.5 Thats almost a 100% difference, and i only changed the index, and indexes are very correlated. I hope in the future I will be able to explain this but I don't think so.:eek: :eek: :eek:
I find profit factor a bit misleading as it looks only at close price vs open price, instead looking at what happened with price in between. I have systems that have profit factor of 4+ but MAR is so bad that they can't leveraged up.
That's a fair point, but the system in question has a holding period of less than 2 sessions and I can typically reduce risk to a few ES points within an hour. A typical trade is entered in the morning and exited before midnight ET. But, because I sometimes do hold beyond the close of the day session, I don't use daytrading margin, so that I can hold the entire overnight session if need be. I've looked at the possibility of using daytrading margin for trades entered early enough in the day, but I actually prefer to stick with the current leverage levels, which are plenty.
And did the profit factors stay approximately similar if/when you went live? I have considered adding the NQ to my trading, but the way my system works, I doubt that the NQ would give me any signals above and beyond what the ES gives me. One of these days, I will start paper-trading the NQ to test that hypothesis.
Your statement doesn't make any sense to me at all. Profit factor is calculated as the ratio of the sum of all winning trades over the sum of all losing trades. Those are closed trades. What happened in between makes no difference and it is not of any concern. PF deals with realized gain and losses only. You probably mean something else.
I took him to be talking about swings within a trade and how they might impact your ability to lever up if the trade lasted a while. I might have confidence that the trade will ultimately swing in my favor (leading to a positive profit factor), but I might get a margin call in the meantime, which would be something I'd want to avoid. That's how I took his comment, anyway.