EPFR FX Allocations: Getting a jump on the yen
Two of the first questions that investors and fund managers ask when confronted with a new data set are, “Does it add value?” and “How do I unlock this value.” Among those that will be available in 4Q19 is EPFR’s FX Allocations data set. In this piece, we look at one quantitative approach using this data set and highlight some promising initial results.
The goal of our approach is to predict G10 currency movements. The G10 includes the US (US$), Switzerland (SWF), Belgium, France, Germany, Italy, the Netherlands (Euro), Japan (JPY), Canada (CN$), Sweden (SEK) and the UK (GBP). To unlock FX signal from this new data set we constructed a flow-momentum model, in the style of Srimurthy, Shen & Smalbach.
EPFR’s FX Allocations data set is based on the FX positioning of over 500 funds, managing over $400 billion in assets, that EPFR collects each month. The numbers are self-reported, sourced directly from the funds themselves. Each asset in the fund is denominated in a specific currency. These weights are then aggregated to the currency level to arrive at FX allocations.
In our preliminary research, we have combined the FX Allocations with EPFR’s weekly flow data to construct a Srimurthy/Shen/Smalbach-style flow-momentum indicator. Each week, for each fund, for each currency, we multiplied net flows into or out of the fund by that fund’s allocation to the given currency. Summing across all funds results in total flows for the given currency from all funds tracked by EPFR during that period.
Repeating this procedure with starting assets, instead of fund flow, yields the total assets denominated in that currency held by these same funds. Scaling the total flows by the assets denominated in each currency yields a weekly flow-percentage number. Following the standard Srimurthy, Shen & Smalbach methodology, we use four-week flow percentage as our flow-momentum indicator. This is merely weekly flow percentage compounded over the latest four weeks.
To begin with, we looked at whether this model had predictive power. We looked at G10 currencies over the four full calendar years for which we had complete data, 2015 through 2018, and looked at the results for a variety of holding periods. Each period, we ranked G10 currencies into five buckets -- or quintiles -- with the first and fifth quintiles scoring the highest and lowest respectively on four-week flow momentum. The table below reports average annualized forward return to each quintile in excess of the average return of G10 currencies for the six holding periods we investigated. In addition, the table lists the annualized quintile spreads, the difference between the top and bottom quintile and the associated Sharpe ratio.
As you can see from the table, for G10 currencies over the calendar years 2015 through 2018 this four-week flow-momentum model has performed. It has a positive quintile spread regardless of the forward horizon used although, not surprisingly, quintile spreads, decline as the holding period lengthens, as do the associated Sharpe ratios.
Now that we’ve established this model works in general, what happens when you use it to trade a specific currency?
In the case of the Japanese Yen, analysis of flows in 2016 highlighted a particularly profitable opportunity flagged by this model. The percentage flow into the yen over the four weeks to Wednesday, 27th January 2016, was 4.5%. This sufficed to drag the yen into the top quintile on flow momentum from the second quintile, the week before. This information, known by the evening of Thursday 28th January, could have been used to trade late on Friday, 29th January 2016. Over the following week, the yen appreciated 5%. The week after that, it appreciated almost another four percent.
This is just one approach to utilizing EPFR’s FX Allocations data set using quantitative models. As is generally the case, more and better ideas will emerge as clients start to work with it.
For more insight subscribe here.