I recently visited the USA. I managed to visit a number of different cities. Whilst I was there primarily for work, I managed to squeeze in a few days to look around the various cities I visited. I visited Philadelphia for the first time, and I saw Independence Hall and the Liberty Bell, both of which I strongly recommend you see. I’m a big foodie (as I’m sure you may have guessed by reading this column!), so I took the opportunity to try out a lot of food stalls and restaurants. I had a Philly Cheesesteak whilst in Philadelphia (where else?!), which I’d been wanted to try for a while. I also ate the Beyond Meat burger, which I have to admit was very good, although it fails to usurp the place of a real cheeseburger in the Pantheon of my greatest food hits.
The thing with food, is that sure, you can go to very fancy restaurants which have very unusual food, but it’s not necessarily the case it’ll be great. Just because food is unusual doesn’t mean it’s automatically good. If you give me the choice between a decent cheeseburger and some exorbitantly priced dish full of rare ingredients, I suspect I’ll derive more happiness from the cheeseburger! You need to actually try food before judging it, and not use it’s price or unusualness as the only criteria.
With data used by financial market participants there’s always the same quandary! Should unusual alternative datasets provide addition value for trading (compared to common datasets)? Theoretically the answer is yes, however, in practice it will require a lot detective work. Furthermore, the answer will vary between market participants and their use cases. I’ve worked in the area of alternative data for a number of years, and am currently co-authoring The Book of Alternative Data with Alex Denev, and the book will be published by Wiley in early 2020.
What’s very apparent is that equities seems to grab all the limelight when it comes to alternative datasets. What about if you’re mainly focused on FX, like myself? It can be trickier to use alternative datasets in FX, because there isn’t always the obvious mapping between an entity and the tradable asset. Sometimes, for example, data or news related to the macro economic situation is very relevant for FX, yet it might not be obvious, unless you are an experienced market practitioner.
One advantage though is that because alternative datasets aren’t as commonly used in FX, there is less likelihood of data “crowding”, where too many people are trying to leverage an alt data signal, so the alpha decays. This has happened to some alternative datasets in equities, which are now much more common.
So which alternative datasets can be useful for FX (some of which I’ve used!)? One new dataset is that from CLS, which contains comprehensive hourly FX volume data and also FX flow data for price takers. I recently published a research paper on the CLS data together with some historical results on using it to generate FX trading signals. There are also a number of news datasets such as Bloomberg News, which can be used to trade FX, if you use your domain knowledge to identify the right news articles. My research paper on using Bloomberg News to trade FX is here. There are also a number of long term capital flow datasets available from Exante Data.
Cuemacro also has our own dataset for measuring the sentiment of Fed communications which uses NLP and another which estimates US nonfarm payrolls using a Twitter based model. We also have a datasets which estimates NDF flows (feel free to contact me if you’d like a demo of any of these).
It is possible to try to aggregate a lot of the equities datasets to give macroeconomic signals (eg. for retail sales). There also the PriceStats dataset which gives inflation estimates using web scraped prices from eCommerce sites.
Obviously, this isn’t a full description of all the alternative datasets which can be used to trade FX, but it should hopefully get you started. If you need advice on how to manage your own FX datasets, Cuemacro can help there too.