You have made an admirable start, including the statistics for analyzing your back testing. I use a four step process that ends with trading with serious money. The names I use comes from my engineering and manufacturing background and I believe it offers some insight into the purpose of the steps Computer Simulation - Back Testing Prototyping - Forward Testing AKA Paper Trading Pre-production - Small money trading Production - Serious money trading If you have not done so yet, I would begin prototype testing (forward testing) with a robot. This is one step nearer to real world trading and may give you some insights that back testing may not provide because of limitations in the data.
If you are using limit orders then you should consider that only a certain percentage of orders will get filled. If you are trading a super liquid market with tight spreads, then it might be better even to pay the spread. But I agree with what most have said. Commissions and slippage will most likely bring your profit factor even closer to 1 if you decide to use market orders. Commissions and the (possibly vastly) reduce number of trades will reduce your overall profit and if you decide to use only limit orders. As to how you will be affected by using limit orders, there's no way to really know unless you have a bash with the smallest possible account size, so I would recommend that if you're happy with results after commissions are factored in. But I agree with the other recent comments, if you're starting at 14, you're miles ahead of the game I was happy with myself, starting at 22, but you may me look like a veteran
Also 24 hour results tend to be unreliable... You could be posting "profits" in dead-of-night illiquid periods... When few actual trades take place... I'll bet you do worse 8:00 to 16:00 than the rest. This kind of thing may be great for a 14 year old... But I could data dredge something similar in about 20 hours work. Without getting too deep, this does not require ANY meaningful expertise, sorry... I was doing the same kind of fantasy quant stuff in my teens... And the one thing you guys will learn about Real Life... Is that you will get absolutely nowhere without becoming an Expert in something... And it takes 10,000 hours = 5 years to become an Expert in anything. Or look at this in reverse... If 20 hours of simple data dredging was profitable... There would be no need to pay... Experienced Pro Traders or Algo Engineers big 6 figure salaries.
Curve fitting is when you over-optimize the input parameters of a trading strategy to the point where you maximize profits, but the actual results won't produce a winning strategy in real life. You can over-optimize any strategy to produce fantastic results. Take any simple MACD crossover on NT, and optimize the results based on various EMAs. You'll find that you will probably have a set of input parameters that will produce fantastic results. That's because you are optimizing your results for past results, including the noise. This will produce great results for the past, however, going forward, because your strategy was optimized for the entire dataset, including noise, it likely won't be applicablein the future, and you'll likely have terrible real-life performance. For a strategy to work, it needs to work on a genuine, reproducible behavioral trait of the markets. That means that it needs to not only work on historical data, but data that you've never seen before. If you used NT backtesting and optimization to generate your strategy, then it likely is unsuccessful strategy. What you should do is take a subset of the data, optimize it for that subset, and then apply the strategy to the rest of the data, and see if the returns you see on the subset work across the entire dataset. If so, then you *might* have a good strategy. Then, you need to forward test it against new, real data to see if it still hold.
Julian, If you want to avoid curve fitting you need to design models that are very robust. Meaning that for a given filter your performance does not dramatically change from excellent to poor. For example if you use a 10 tick stop, and your results are great, if you use an 11 tick stop, are your results complete crap? A good way to do this is to plot your attributes against PNL, and see if the curve is nicely rounded vs very choppy.
jedwards got it right on. i use tradestation and i can almost use any of their sample strategies and over optimize it to show me fantastic returns in back testing. you put in 'tick' as back-testing resolution and they all collapse. also you said there is no profit target or stop order in your strategy. that's just insane from risk perspective.
I don't understand why most of you are wasting your time trying to understand what the merits are of a system with profit factors of 1.18 and 1.2. This is basically a worthless system. There is no need to over-analyze things. The kid should be told that any system with a profit factor less that 1.5 is a random system unless he wants to learn that the hard way and he will if he doen't listen, no doubt about that, he will be crying over this lost college tuition money because someone is waiting to take it from him and he will be glad to because his system has an infinite profit factor, i.e. he is a market maker.
Try being more selective in your trade selection. Aim for more than $50 a trade. That is missing one tick in and one tick out. That way you have a better chance of being profitable.
Yes, time to start forward testing this strategy using a paper trading account. I use IB TWS Paper trading account for this. Fills and slippage are simulated. While simulating for 1-2 months don't change a thing in your parameters. You can ofc backtest different things but keep the paper trading as stable as possible. After 2 months analyse your trades and see if optimization is possible. Nice job for a 14yo! Keep it up!