Workflow in algorithmic strategies

Question

Sorry for the basic question, I'm trying to educate myself on algorithmic strategies.

Just to see how it works, my idea is to create a simple moving average strategy.

Let us suppose I have $N$ observations of the price of an asset $P_t$

I define the simple moving average of $P_t$ at time $t$

$$\text{SMA}_t^{(n)} = {1\over n}\sum_{k=t-n+1}^t P_k~,$$ where $n$ is the number of prices included in the average.

Now the trading signals are generated by these two trading rules:

• $\text{BUY}_t$ when $P_t < \text{SMA}_t^{(n)}$
• $\text{SELL}_t$ when $P_t > \text{SMA}_t^{(n)}$

Every time I have a signal I buy/sell everything I can/have, in other words I maximize the volume of my operations.

My question is how to test this strategy? I tried with the $SP500$ index and I found very good results in terms of $\%$ returns, but I think my approach is misleading since I should compare it with a buy and hold strategy, considered that this index has only grown in the period that I considered...I also read that simple moving average is good when the market is mean-reverting, and this makes sense to me... What are more appropriate way to test it with real-world data? Also, can you suggest the next steps to make it more realistic?

Thanks in advance

Answer 1

"What are more appropriate way to test it with real-world data? Also, can you suggest the next steps to make it more realistic?"

Pick a tradable instrument (e.g. SPY rather than S&P500), eliminate look-ahead bias (e.g. trade at the next day's open rather than the close), take into account transaction costs and financing costs, estimate market impact/slippage...I'd guess there are some blog posts and books aimed at retail traders that cover these basic ideas, e.g. from Rob Carver [EDIT: I think Ernie Chan's books might also be of use/interest to you. I have no affiliation with either Carver or Chan]