If there’s one thing I’ve done a lot of over the last few years, reading would be it. I’ve devoted a great deal of time to devouring any material that I thought might give me an edge in my trading – textbooks, academic papers, blog articles, training courses, lecture notes, conference presentations…anything and everything I could get my hands on.
I was browsing the folder called “Trading Library” in my Dropbox this week and found myself fondly recalling how certain authors had contributed to my knowledge. Sadly, I also found myself shaking my head at the time I’d wasted sifting through useless material, of which there is an unfortunately abundant supply. This inspired me to write a post that catalogs the material that I feel provided me with the most valuable knowledge and insight into topics related to algorithmic trading. I hope it saves you some time and money.
I should note that this list is far from exhaustive. Above I mentioned my Dropbox folder called “Trading Library”. Well, its largest sub-folder, by a long stretch, is the one labelled “Yet to Read”. As I work through that folder and come across other useful material, I’ll update this post accordingly.
Of course, none of the material I list here provides anything like a “holy grail” capable of leading the trader to overnight riches. In order to make it onto the list, the book or paper simply needed to present an idea, technique or insight that helped me in some way on my journey.
The material is not listed in any particular order of importance or preference and is organised into the following categories:
- Books about algorithmic and/or quantitative trading
- Textbooks which cover important foundation concepts
- Books about machine learning
- Academic papers
- Courses and related material
If you have any other recommendations, I’d love to hear about them in the comments.
Books about algorithmic/quantitative trading
Aimed at retail traders, this is a highly practical guide to succeeding at the business of quantitative trading. Includes oft-neglected topics such as sourcing good data, choosing a development environment, advice on working as a proprietary trader, and the importance of good execution. Also introduces a number of approaches to strategy development and includes a coverage of risk management.
Delves into the specifics of momentum and mean-reversion trading strategies. Probably the most comprehensive coverage of the latter that exists in the public domain.
Something of a call to arms for the technical trading community to adopt a scientific approach to the markets. More than that, it provides some practical tips on how to do it following the work of White (2000). I particularly enjoyed learning about the various cognitive biases to which the human brain is subject and how to account for them.
A practical introduction to leveraging R for quantitative analysis of financial markets. A great way to learn some applied R skills as well as some quant finance concepts at the same time. Includes an introduction to using R for the following: statistics and probability, spread trading, handling high-frequency data, options modelling, backtesting and optimization.
Fitschen doesn’t break a lot of new ground with this book, however for me it was worth reading for its approach to quantifying the differences between various markets and how that information can be used to practically develop a trading system. While it’s certainly not the last word on the topic, Fitschen’s treatment of it got me thinking in a different way that resulted in tangible improvements to my trading.
A compilation of papers describing the basis of and results for various trading strategies. Ideas presented include regression analysis, cointegration, models for the term structure of interest rates, recurrent neural networks for forecasting currency volatility, using machine learning for modelling credit risk, regime switching, among others.
The main utility of this contribution is its treatment of digital signal processing methods for financial applications. Such methods enable the construction of high-performance filters and smoothers which are generally more responsive and efficient than the typical moving average filters that are ubiquitous in finance.
A useful starting point for traders from a technical analysis background who are interested in making their systems development process more robust. Introduces optimization, curve fitting, monte carlo analysis and trend bias.
Pardo provides a useful introduction to the use of simulation to evaluate and optimize trading systems. This is another beginner level text that includes a useful treatment of simulation accuracy, statistical measures of performance, optimization, curve fitting and walk-forward analysis.
This is compelling reading that is a must for systems traders. Vince provides a mathematical treatment of money management including ideas borrowed from gambling, key probability concepts, reinvestment, optimal growth, Kelly betting, modern portfolio theory and the ‘leverage space’ portfolio.
This is the go-to reference for the maths of financial time series, covering linear and non-linear time series models, high frequency data analysis, continuous time models (Ito’s Lemma, Black-Scholes etc), multi-variate models, Kalman filters models and Markov chain models.
Spiegel, M and Stephens, L. (2008). Schaum’s Outline of Theory and Problems of Statistics, McGraw-Hill.
Having a good working knowledge of statistics makes strategy research and development efficient and robust. This volume is a great place to start and a handy reference. The chapter on time series analysis is a good primer, but is worth supplementing with material which has a specific focus on finance, such as Tsay (2010).
Hyndman, R. and Athanasopoulos, G. (2016). Forecasting: Principles and Practice, otexts. https://www.otexts.org/fpp
A handy online reference for applied time series forecasting using R.
Machine learning books
If I could only recommend one text on machine learning, this would be it. However, Applied Predictive Modeling encompasses far more than just machine learning – it also delves deeply into the process of robust development and model evaluation. Includes many examples covering a wide array of learning algorithms and modeling scenarios. Examples are written in R and make use of the author’s caret package, which is also highly recommended for streamlining the modeling process.
A broad overview of machine learning applied to several example problems. Specific learning algorithms introduced include k-Nearest Neighbors, naive Bayes, decision trees, linear regression, regression trees, neural networks, support vector machines and k-means clustering. Start here if you have no machine learning experience for a gentle but practical introduction.
A collection of papers with some innovative use cases of machine learning for financial applications. I especially like the idea of using hybrid ARMA-Neural Network models for improving forecasting systems.
A very readable introduction to using recent developments in machine learning to model complex data sets. In terms of algorithms, the book covers all the usual suspects: linear and non-linear regression, logistic regression, regularization methods, tree-based models, support vector machines, PCA and k-means clustering. There is also an introduction to resampling methods including cross-validation techniques.
The companion text to An Introduction to Statistical Learning, this volume covers the same material (and more) but is intended for an audience with a relatively strong mathematics background and is a much more rigorous, technical treatment of the subject. Includes additional topics such as neural networks, random forests, and ensemble methods.
First and foremost a software manual for Aronson’s development platform, TSSB, this book nevertheless contains numerous practical insights into leveraging the power of machine learning for financial prediction.
A comprehensive compilation of the authors’ experiments with neural networks applied to the foreign exchange markets. The volume covers a range of experimental setups, network architectures and hybrid linear/non-linear models. There is a lot of material here, but I found it a useful source of ideas. All the more intriguing since the research was published before the recent surge in interest in deep learning architectures.
This paper is widely considered strong evidence for the utility of analysis of historical price data (in particular, the ‘momentum’ of an asset) in predicting future returns, at least in the equities markets. Interesting conclusions regarding the relationship between the seemingly-contradictory factors of ‘value’ and ‘momentum’ in returns forecasting.
As the name suggests, an exploration of the existence of momentum in the currency markets. The authors conclude that momentum does exist, depending on the underlying autocorrelation of the returns series.
This paper explores the interactions between momentum and mean reversion , ultimately expressing excess returns as a combination of momentum and mean reversion terms. This approach is innovative in that it differs from the more common binary regime-switching model.
A method for accounting for data mining bias in trading systems development. This paper was popularized in David Aronson’s Evidence-Based Technical Analysis in 2007.
This assessment of technical trading rules and strategies concluded that there is a case for the existence of profitable trading strategies when accounting for data mining bias using the method of White (2000).
Abril, I and Sugiyama, M. (2013). Winning the Kaggle Algorithmic Trading Challenge with the Composition of Many Models and Feature Engineering, IEICE Transactions on Information and Systems, vol.E96-D, no.3, pp.742-745.
An empirical model for predicting the short-term response of the bid and ask order book following a liquidity shock. I found this a useful example of a modelling approach using machine learning techniques applied to finance.
Domingos provides 12 key insights or ‘rules of thumb’ to keep in mind when applying machine learning to a problem. These are incredibly useful from many perspectives – setting expectations of the modeler and guiding the modeling process to name two. Some of my favorite quotes from the paper:
“Generalizing correctly becomes exponentially harder as the dimensionality (number of features) of the examples grows, because a fixed-size training set covers a dwindling fraction of the input space.”
“At the end of the day, some machine learning projects succeed and some fail. What makes the difference? Easily the most important factor is the features used.”
“As a rule of thumb, a dumb algorithm with lots and lots of data beats a clever one with modest amounts of it.”
An interesting idea that I applied in one of my early blog posts.
The authors use an autoencoder to extract features for input to a deep learning neural network architecture to discover an edge in US equities markets.
The authors used a Hidden Markov Model with two states to account for the severe drawdowns associated with a momentum strategy.
A framework for developing trading systems based on learning algorithms.
Jiang, Z. (2016). Currency Returns in Different Time Zones
An exploration of the variation in exchange rate dynamics across different time zones and an interesting explanatory hypothesis.
Courses and course materials
Andrew Ng’s Coursera machine learning course. A fantastic resource that breaks down complex topics into a series of immensely understandable lessons. Lays a strong foundation for further study in the field.