Michael's blog

Photo by Aaron Burden on Unsplash

In the last post, we learned the basics of performing the pair trading strategy and using cointegration as a method to identify the potential tradable stocks pair. All the theories and the math formulas are so seemingly promising and convincing enough for us to believe it’s a profitable and stable trading strategy. But is it? In order to test and check the profitability and effectiveness of this strategy, we need to backtest this trading strategy to simulate real-world scenarios.

Cointegration is a statistical technique to find out whether a time series closely follows the movement of the other time series. Therefore, it becomes an important technique in the pair trading strategy for us to determine the right stock pair to trade with. In this post, we’re going to see why traders prefer using the cointegration test over the correlation test in pair trading, and whether the cointegration test results can boost our trading performance.

In the last post in this series, we’re going to look at some questions that I discovered while working on connecting to Interactive Broker API. Some of them are due to the obscurity of the configuration and hard to find the right place to configure them, and some of them would need the extra tool to resolve. I put all of them down into one post and share it with you.

Now we come to the third part of this series. In this post, I’m going to show you how I design and build my local database to store IBKR trades and other necessary information for generating meaningful indicators to review our strategy performance.

This is the second part of the Set Up Trading API Template In Python. We’re going to focus on implementing the rest of the functions in our Interactive Broker class.

Building your trading strategy to connect to a broker with the broker’s proprietary API is always dreadful. There are tones of API documentation to read, tones of trial-and-error tests to conduct, and tones of unknown causes and bugs that fail your API test. In this post, I’m going to demonstrate my MVP API template to get my trading strategies to work, so that you can build your own in a way that makes your trading strategies work as well.

Photo by Sasun Bughdaryan on Unsplash

We’ve been talking too much about the attack side of quantitative trading, such as momentum, mean reversion, and ML. These strategies aim to outperform the benchmark/index by adding your personal points of view to the trading strategies. Beating the benchmark becomes the only goal when playing the offense. What about defense? After reading Introduction to CPPI – Constant Proportion Portfolio Insurance, I started to feel that I can’t agree more with the idea of “The best defense is a good offense” once said by Sun Tzu, a Chinese military general, a strategist, and a philosopher. What does defense mean in the field of quantitative trading? Does defense mean we strive not to lose money and then nothing else worth doing? Maybe talking about the CPPI strategy would give us a better picture of what actually defense means to the traders. Let’s now have a look at how to approach the other side of trading.

The Triple barrier method and meta-labeling technique were together introduced in the book Advances in Financial Machine Learning by Marcos Lopez De Prado. It seems that the combination of these two tools makes a great pair to either stabilize or further increase your portfolio growth. In this post, I’m going to quote my old research result (here) from last time as the fundamental strategy benchmark, and apply these two techniques to see what beneficial impact we could bring to this strategy.

When it comes to using machine learning algorithm to pick the stocks that are most likely to produce a good return, it is similar to seeking the opinion of an investment consultant. However, it can be unsettling to make your investment decision after listening to just one consultant. Now is the moment to get second opinions and hire more investment advisors to make sure the investment concept is reliable, doable, and profitable.

The same principle that you consult other machine learning algorithms to confirm the predictions made by these models are applied in ensemble learning. When you have collected all of the final data from these models, you may take your time relaxing in your nice chair like a big boss, analyzing the results, and making your important and sacred decision.

Photo by Brett Jordan on Unsplash

From the previous article, we’ve learned several indicators that we can calculate and use to evaluate the performances of your algorithm trading strategies. Given these indicators, we’re able to see how we can further polish our strategy and make it more seemingly profitable. Therefore, let’s work on our features to see how we can improve our machine learning trading strategy.