Last week, we looked at simple data analysis techniques to test for persistence. But we only looked at a feature that is measured over a single day – the absolute range. Such a feature makes it easy to test persistence because you don’t have the problem of overlapping data. Each data point is entirely self-contained and shares no information with the previous data point. Compare this with a feature that’s calculated over a number of days, for example volatility estimated as the standard deviation of the previous 30 days’ returns. Today’s estimate only differs from tomorrow’s estimate by a si ..read more

An assumption we often make in trading research is that the future will be at least a little like the past. I see a lot of beginners making this assumption implicitly without recognising that they’re making it or thinking about whether it’s reasonable to do so. That’s a mistake. If you are making this assumption, you need to be aware of it and you need to have confidence that it’s a good assumption to be making. There are a couple of ways we can explore whether it’s a good assumption: Reason and eyeball Data analysis In this article, we’ll look into both of these approaches, but we’ll most ..read more

Here’s a thing that I suspect will make money, but that I haven’t yet tested (for reasons that I will explain shortly): Every day, at the start of the trading day, get the SPX straddle price and convert it to an expected SPX price move. Then at the end of the trading day, take the SPX price and calculate if it moved more or less than the straddle implied. Aggregate this over a few days – the simplest way would be to take the average expected move and the average actual move. The trading signal for the next day’s open is as follows: If SPX moved more than was implied by the straddle on aver ..read more

Here at Robot Wealth, we trade with Interactive Brokers (IB) primarily because they offer access to global markets at a reasonable price. In recent times, IB has put some time and effort into upping its tech game, including development of an API for interacting with its desktop trading applications. An application that interacts with IB’s desktop trading applications via the API is essentially a message-handling program. So I want to show you a simple but effective architecture for managing the flow of messages and the operations they trigger. If you can understand this framework, then you c ..read more

Navigating Tradeoffs with Convex Optimisation This is the final article in our recent stat arb series. The previous articles are linked below: A short take on stat arb trading in the real world A general approach for exploiting stat arb alphas Ideas for crypto stat arb features Quantifying and combining crypto alphas A simple and effective way to manage turnover and not get killed by costs How to model features as expected returns Building intuition for trading with convex optimisation with CVXR So far in this series, we have: Introduced a general approach for doing stat arb Brainstormed s ..read more

This article continues our recent stat arb series. The previous articles are linked below: A short take on stat arb trading in the real world A general approach for exploiting stat arb alphas Ideas for crypto stat arb features Quantifying and combining crypto alphas A simple and effective way to manage turnover and not get killed by costs How to model features as expected returns Next, we’ll build some intuition for using convex optimisation to manage real-world trade-offs in execution. Imagine you’ve quantified some signals that are predictive of next-day returns. You’ve done the grunt wo ..read more

Modeling features as expected returns can be a useful way to develop trading strategies, but it requires some care. The main advantage is that it directly aligns with the objective of predicting and capitalising on future returns. This can make optimisation and implementation more intuitive. It also facilitates direct comparison between features and provides a common framework for incorporating new signals or reassessing existing ones. Finally, expected return models can be integrated with common risk models such as covariance estimates, and enable a direct comparison with trading costs, ope ..read more

Every time we trade, we incur a cost. We pay a commission to the exchange or broker, we cross spreads, and we might even have market impact to contend with. A common issue in quant trading is to find an edge, only to discover that if you executed it naively, you’d get killed with costs. In this article, I’ll show you an example of using a simple heuristic that helps you do an optimal amount of trading to realise your edge in the face of costs. A heuristic is a simple rule or shortcut to help make decisions. And this one doesn’t involve any fancy math or portfolio optimisation and is quite ef ..read more

This article continues our recent articles on stat arb: A short take on stat arb trading in the real world A general approach for exploiting stat arb alphas In this article, I’ll brainstorm some ideas for predictive features that you could potentially use in a crypto stat arb model. The ideas draw insights from recent discussions and market observations, but of course, you should do your own research. In future articles, I’ll pick some of these features and seek to quantify and combine them, and finally demonstrate how to navigate the return-risk trade-offs using heuristics and explicit opt ..read more

Last week, I wrote a short article about statistical arbitrage trading in the real world. Statistical arbitrage is a well-understood concept: find pairs or baskets of assets you expect to move together, wait for them to diverge, and bet on them converging again. Simple enough. But making it work, especially at scale, is a little more complicated. A somewhat old-school approach takes pairs of correlated assets and trades them long-short together. This is the classic pairs trade. But we don’t have to constrain ourselves to this approach. And when we consider a more general approach, we find exp ..read more