Here is an article from Quora which may give you some insights into system development.

Above all else, what you should consider first and foremost is that this backtest means very, very little, on its own.

I can't possibly overemphasize the importance of this. Here's an interesting/revealing little fact that will hopefully make you think twice before putting any implicit faith in these backtest results. .

The first thing I did when I dove headfirst into algorithmic trading was to try to test, objectively and on a massive scale, several foundational concepts/truths/ideas. One of the most important of these was 'Just how predictive is a strong backtest of out-of-sample, forward-walk success?'. I did this by making use of the millions of rows in our sprawling database, each of which represents an individual algorithmic strategy based on fundamentally unique logic. The data-obsessed side of me wants to write a novel about the discoveries therein, but I'm going to force myself to keep this short. . . what I did, in effect, was to then test the cream of the crop of all of these against both historical out-of-sample data, as well as live-walk data (real-time trading in live markets) over the next year or so. These strategies were a very wide sampling of instruments, entering both 'long' and 'short' equally within each instrument, and they contained a wide variety of core logic. . . a very wide ranging, objective sampling.

I created two mock-portfolios, one containing the top 1-2% of all of the strategies throughout the database (where 'top' is determined by a singular generic strength score created by using a relatively simple calculation considering earnings relative to losses, drawdown, and sharpe ratio) investing one-contract-per-trade across the board, the other a 'weighted' portfolio, that invested more heavily or lightly according to the strength score.

The results? The first group just barely broke even when their results were taken as a cumulative whole over the entire out-of-sample period, after considering slippage and commissions. The second 'weighted' group did just slightly better, approximately a 1.03 'profit factor' (which is a simple earnings-divided-by-losses calculation), which would be justifiably considered breakeven in most circles.

The very best of the best backtest/optimization results one could hope for (we have the software and processing power to produce stunning on-paper results), and the walk forward trading results were approximately random.

This was one of the most valuable lessons I've learned, and I'm thankful I decided to perform these tests early on, there's no telling how much heartache has been avoided.

It's immensely tempting to look at the results of a candidate for an automated trading strategy, especially one that has traded frequently over a long period of time, and naturally expect that the results will continue to be strong. . not necessarily mirroring the historical results (expecting this is insanity), but at least holding on to enough 'edge' to make a tidy profit. In practice, this is not the case.

Much value CAN be discerned from backtest results, but it's far more complex and difficult to do so efficiently/accurately/consistently than you might think. It has more to do with the interplay of statistical data points, than it does with any singular data point (such as 'net profit'). Furthermore, far more value can be gleaned through several other creative tests, as opposed to a simple backtest where you then eyeball the resulting stats. . . such as testing the core logic of the strategy on correlated instruments, or over different hour-ranges (if day trading), etc. I can't elaborate too much here, as this quickly enters secret-sauce/proprietary-info territory, but I say these things to illustrate the point that while a backtest should never be trusted implicitly, it can be trusted to likely be better-than-average, if you know where and how to look.

Even still, this is only a starting point at best, as any backtest should always be. . . a starting point and nothing more. You'll want to watch it trade live (on a simulation account!) for a fair amount of time, to be sure that it performs roughly as you'd expect based on the historical record. . pay extra attention to the size of the largest win, loss, its win% or ratio of win size to loss size, its drawdown, etc. . . if any of these deviates too far from your historical results, chances are good that something is amiss.

Also, always consider the general trend of the market/instrument, both during its historical period, as well as its live trading results. If it's an automated strategy that only takes long entries, and its brief live-walk period in which you've actually monitored the trades has been one where the instrument you're trading has shot to the moon, its best not to give it too much credit for these returns (the opposite is equally true, a strategy that performs even decently in significantly adverse market conditions is almost always worthy of some attention).

I hope you haven't blown out an account since writing that question. . but if you have, don't give up, virtually all great traders have been there. Learn all you can from it, and if you're still enjoying what you do, you'll look back (likely from a much better/wealthier vantage point!) someday and laugh.

Good luck!