So much has been written lately about the integration of Quantitative Investing and Machine Learning.
Are you wondering about whether to apply ML to your investment process or feel like your team isn’t progressing fast enough? We hope that our new blog series on applying Machine Learning algorithms will help you.
Over the next several blogs, we will:
- list the steps for getting started with using Machine Learning
- dive into more detail on the getting started steps
- discuss some of the more popular prediction algorithms
- show how to integrate these algorithms into a predictive model
- share best practices for building and reviewing the efficacy of Machine Learning Factors and Models
At a high level, Machine Learning is a category within Artificial Intelligence where patterns are found in “training data” and those “models” are then used to make predictions. In Quant Finance, we can apply these algorithms towards objectives like predicting stock prices and optimizing investment portfolios. Using ML is not all or nothing, it can be used as some of many inputs into your investment process.
Here are a few steps to help you get started:
- Select a ML Platform: Every day, more commercial investment products support the application of ML algorithms into their analytics toolsets. These applications generally insulate users from the time needed to understand the details of applying each technique. If you prefer rolling up your sleeves and programming, most ML algorithms have libraries built for the popular quant languages, like Python, R, and MATLAB. At Scientific Financial Systems, we’ve heavily used all three of those languages, but have settled on python for its openness, comprehensive libraries, easiness to learn, and easy integration with other technologies. Our Quotient™ tools allow users to simplify the integrate of ML algorithms into their factors and models using a user-friendly graphical interface and/or python coding.
- Understand your data: Constructing a factor starts with an investment thesis and actualizing that thesis requires relevant data. Just like when building “regular” factors, ML factors require that you fully understand the scope and quality of your data. This includes understanding the definition of columns, identifying missing or erroneous data, and checking the data for bias. Various statistical methods and graphics within Quotient™ can help.
- Select the ML algorithm that matches your thesis: There are MANY machine learning algorithms from which to choose, and that list continues to grow. You will see names like linear regression, decision trees, random forests, and neural networks. Quotient™, by example, currently supports over 50 ML algorithms and that list continues to grow. It’s important to choose an algorithm that reflects your thesis and not to data mine through the running of every algorithm looking for the best result. You will also want to optimize your selected algorithm using techniques such as hold-back periods or cross-validation.
- Build and train your model: Once you’ve decided on an algorithm, it’s time to build and train the model using your data. This involves selecting the appropriate parameters, splitting the data into training and test sets, and evaluating the model’s performance. There is a practiced art on how to perform this step properly; especially when it comes to tuning without introducing bias. Like other tools, Quotient™ will simplifies the building of a model by displaying the parameters associated with each ML algorithm. At Scientific Financial Systems our team of experienced Quants and Data Scientists work with clients and suggest ways to limit such bias.
- Apply model to live data: This is the part you’ve been working towards! Once your model has been trained and you are comfortable with the results, you can use it to make live predictions. This involves feeding the new data into the model and interpreting the results. The best tools will allow you to use the model you used to test history on your live data – without modification. This reduces the chance of a mistake when migrating from research to production.
- Continuously monitor and improve the model: Using investment finance models is like being in an arms race. Your opponents are always advancing their technology, and what had superior predictability one day suddenly seems as effective as a dart board. Remember that Machine Learning models are not static and must be continuously monitored and improved. This will mean periodically re-evaluating your model’s performance and testing with new data sets and intelligent tweaks of your parameters.
To summarize, getting started with Machine Learning in quantitative finance involves understanding the basics of machine learning, and a process for building out your architecture and designing, deploying, and reviewing your models. In our second blog of this series, we will jump ahead and review some of the more popular and effective Machine Learning algorithms used by today’s Quants.