stocks

Our Statistical Process for Stock Market Research

Stock Market Guides is not a financial advisor. Our content is strictly educational and should not be considered financial advice.

Everything we do here at Stock Market Guides is based on data and statistics. For every trading strategy we feature in our service, it had to first pass a series of rigorous research tests.

Here are some of the cornerstones of our statistical process:

 

Decades of Data

For a trading strategy to even be considered for use in our service, it had to be tested across decades of historical stock price data. The strategies we feature aren’t ones that did well in backtests for just a year or two. These are strategies that really stand up to the test of time.

A key reason we want to review the trading strategy across such a long period of time is to have information about how it performs in a wide variety of market conditions.

Another reason is to make sure the results of our backtests are statistically significant. In other words, we want a large amount of data to increase our confidence in the backtested edge.

Here’s a helpful way to look at it. Imagine in baseball if the MVP of the league has a game where he doesn’t get on base in four at bats. If you looked just at that small set of data (4 data points), you might conclude this guy wasn’t very good.

But that wouldn’t be an accurate conclusion in this example. So by looking across a much larger set of data, such as a full season in baseball, you get something that is statistically significant and therefore a better representation of the truth.

 

Raw Edge

We look for trading strategies that reflect a raw statistical edge in backtests. This means that we don’t want strategies that only work with one single magical set of variables. We instead want strategies that are robust and performed well in backtests using a variety of different variables.

For example, let’s say you run a backtest for a trade strategy, but the only way it gets good results is if you set the profit target at exactly 4.3% and the stop loss at exactly -1.9%. If the strategy doesn’t work well at other profit targets or stop losses, then the trade doesn’t have raw edge. And if it doesn’t have raw edge, then it likely isn’t an edge at all.

We use trade strategies where even if you change the variables to some degree, you still see profitable results in backtests. In other words, when you find a strategy with raw edge, it means that it captures a period in time where a stock might have higher odds of going one direction than another, regardless of when you exit the trade. That’s what we’re looking for: a robust, raw edge for each strategy.

 

Statistical Tests

When we run a backtest, we don’t focus exclusively on the net profit of the strategy. We care about how much profit it generates relative to the risk taken. Quantifying that risk is a huge part of our statistical process.

We pay special attention to drawdowns that the trading strategy had in backtests and in live trading.

We like to perform standard deviation tests on our live trades to make sure they fall within the expected scope of the backtest performance. In other words, we want to make sure that if we’re going through a drawdown, the performance is still within the bounds of what the backtest exhibited.

And when we perform that standard deviation test, we’re doing it as part of an out of sample test. This simply means that we are looking at live trades as they happen to make sure they perform similarly to the historical backtest. This is another form of confirmation that the strategy might have a legitimate edge.

We also perform Monte Carlo tests on our research results. Although it’s helpful to see what the drawdowns were on the actual backtest itself, we want to also see what might have happened if the sequence of trades were changed.

In other words, what if the trading strategy had the exact same profit potential (average profit per trade and win percentage), but the sequence of trades was different? That’s what we are testing with a Monte Carlo test. And we test 10,000 different variations to see how deep the drawdown may have gone in a worst-case scenario.

It's only after we are satisfied with the results of these tests that we’d consider deploying any given trading strategy in our service. Going through this barrage of quantitative processes improves our odds of giving customers trading strategies that might have a true edge.