I end up buying a lot of books. Inevitably, I end up reading far fewer books than I end up buying. The unread books peer at me from my bookcase, knowing they’ll likely never be read. One book which I bought recently and immediately started reading was the Trading Game by Gary Stevenson, a former STIRT trader. All I would say is that a book about a trading desk is likely to be filled with more revelations than anything written about the life of a researcher in a bank. Perhaps a more researcher focused book would be filled with revelations about discovering INDEX and MATCH functions on Excel, and how much faster they are than using VLOOKUP (believe me, I did find this revelatory when I discovered it). Or perhaps it would devote a few pages to the inevitable discussions about how you are asked to do something which sounds trivial by the sales or trading desks, but is in fact extremely non trivial.
Ok, I’m perhaps being a bit facetious here, after all there is more to doing financial research than exploring the vagaries of Excel or responded to requests from other parts of the business. Over time, you build up a picture of how you think market works, and once you think you know it all, some event occurs which totally dispels that: markets lull you into mistakenly thinking you understand them fully. However, experience does give you a sense of where to look, when deciding what topics to research in market. The difficulty with researching financial markets is that there are countless avenues to explore. There are many datasets out there, there are many types of models you could investigate, many markets etc. Time is limited, so you need to somehow judge what projects are worth exploring. Indeed, it is something we think about a lot at Turnleaf Analytics, where we are forecasting economic data releases, in particular inflation. We continually undertake research to improve our forecasts, and a big part of that is deciding precisely what topics to research. Very broadly there are two choices you need to make when researching financial markets (I suppose you could generalise this to many areas of data science):
- Trying something totally new
- Tweaks to an existing model
Each approach has pros and cons. If you keep trying to research something new, then you won’t have enough time to maintain existing models, whose alpha might diminish over time (particularly with more constrained capacity strategies). Equally, if you just keep working on making small tweaks to an existing model, you might be ignoring fruitful areas of research you haven’t explored. Ideally, I would say you need to balance the two. The key difference between modelling financial markets, is that you have a system where participants are continually changing their behaviour, so a specific edge is unlikely to last forever, which necessitates the need to research continually and not simply rest on your laurels. Indeed, I recently saw a research paper investigating certain patterns in FX which I had been investigating over a decade ago with Brent Donnelly, who currently writes the AM/FX newsletter, a reminder that eventually with enough time folks will catch up! Looking back, I have also realised after many years, there is still value in researching something with a low delta of success. At the very least disproving something that traders use to make trading decisions, can save money. Sure, the high delta ideas have more chance of success, but potentially the payoff could be smaller. You can also try to break down your low delta ideas into a sufficient number of smaller problems.
In a sense you can think of your research process as a portfolio of different ideas with different levels of risk. Just like you risk manage a trading book, you need to risk manage your research process. Allocate some risk to low delta ideas, some to higher delta ideas, have a mixture of totally new models, some tweaks to existing models etc. Having experience can not only help identify areas of research (ie. where to allocate “research capital”), it can also help you estimate the chances of success for all these research ideas. Importantly, work with your team to understand how to allocate capital, so everyone can chip in with ideas. Creating models is a complex process and requires many steps from finding the data, to creating the model itself. It really is a team effort, everyone has a part to play. Just like with managing capital, you never want to overleverage when doing research, otherwise, you will simply end up with multiple research initiatives that start and never finish. You have to be aware of the constraints in terms of research time, and also when to “stop out” of a strategy, and move on to something else.
It isn’t easy knowing where to spend your research time, but having a proper strategy for it is probably the closest thing you can have to a “secret sauce” in financial markets.