The Essence of Trading System Development (Part 1)

Let’s step back and discuss the big picture of how I currently conceptualize trading system development.

In my view, the goal of trading system development is to find viable—not profitable—strategies. Viable means “capable of working successfully,” and while I certainly hope every viable strategy ends up being profitable, system development stops where live trading begins. It may be just semantics and a matter for debate, but I don’t want to hear people claiming to have discovered profitable trading systems that have not actually been traded live. That smacks of optionScam.com since only live trading can reveal whether a system is capable of generating real profits.

I suspect finding strategies that profit in live trading is more difficult than finding viable strategies. I don’t have numbers to support this yet. Either way, I am focused on the latter.

In order to find viable strategies, one needs to have a logical process. I believe I have that now. In fact, I have two mutually exclusive processes that I will talk more about as we go along.

“How to develop trading strategies?” has no single right answer. This seems like the objective function, which I deem to be more subjective in nature. Remember that the development phase aims to give us the confidence required to stick with an unbroken strategy in the face of “normal” drawdowns (as mentioned in the third paragraph of Part 4).

The goal of developing viable strategies is to find a model to fit the signal. A signal is present in both historical price data and future price data. We use historical price data to come up with a model that we hope will match future price data. If this happens, then the equity curve will slope upward to the right and make money. If the model does not fit [whatever ends up as] the future signal, then the equity curve will look ugly and probably lose money.

I try to bias backtests against me since backtested results inevitably turn out to be overly optimistic. I feel strongly that transaction fees (slippage and commission) should always be included (see second and third paragraphs here). Also, for this reason I won’t accept a backtested equity curve that looks fair/good. I want to see excellent because fair/good might end up as poor/fair when traded live.

In trading system development, viable strategies will have a high signal-to-noise ratio (SNR). I mentioned the signal above. Noise is random and therefore different in the past from what it will be in the future. When I find a strategy that backtests well but does not perform going forward, I have created a model that fits noise more than signal. This represents a low SNR. We want the opposite.

I will continue next time.