The other side of the FX Allocations coin
In our previous Quant’s Corner, Bulls in an FX Shop, we discussed how isolating bullish positions in EPFR’s new FX Allocations dataset can generate excess returns. In this one, we explore what happens when we use the new dataset to drive models created to harness EPFR’s country allocations data.
This work was driven by the need to address a frequently asked question: What does FX Allocations offer that utilizing EPFR’s well established Country Allocations data does not?
In the course of our comparative work we isolated a promising contrarian signal that offers a new way to extract value from existing, Country Allocations-based models.
Apples and Oranges
Country allocations are, by definition, a fund’s allocation to each country. But that allocation does not necessarily translate directly into that country's currency.
Take Chinese technology company Ali Baba. A fund investing in Ali Baba’s N shares would be investing in a Chinese stock, and that would be counted as a China Allocation in EPFR's country allocation database. But the actual investment would be in USD, as the stock is trading on NYSE. Hence, the currency risk of the fund also would be in USD.
Emerging markets hard currency bonds are another case in point. A Brazilian sovereign bond denominated in USD would be counted as a Brazil allocation in EPFR's country allocation database. But the actual positioning and the currency risk would be in USD.
To allow clients to discriminate between country allocations and the actual FX allocations of funds, EPFR now gathers the net FX allocation information of funds. The resulting database, FX Allocations, tracks the allocations since 2015 of over 1,800 funds managing assets totalling $2.8 trillion to more than 90 currencies. The data is adjusted to factor out hedging activity.
The FX Allocations universe is currently three-quarters the size of the Country Allocations universe. But a significant amount of cross-border equity funds which are not reporting their country allocations to EPFR, are reporting their FX allocations.
Old model, new fuel
For several years EPFR’s quant team has used cross-border equity funds' allocations to drive a flows-based FX strategy that has generated sustainable returns going back to 2008.
This FX Strategy is rather simple and follows the flow-based indicator created by Srimurthy et al.. By calculating daily flows to each currency, and using the 20-day compounded flows, it ranks currencies in terms of their flows. The strategy goes long in currencies that have received higher flows and goes short in currencies that have received lower flows. The results are tabulated below.
For the purposes of our research, we decided to progressively deform the current country allocations-based FX-flows predictor, step-by-step, until we arrived at a FX-flow predictor based on FX allocations.
The case for the status quo
To make this shift, we tested four permutations of the data that can be gleaned from both datasets. These were:
a. Country Allocations – the standard Srimurthy-Smalbach-Shen-style 20-day flow percentage model based on EPFR’s Country Allocations data.
b. Country common – as for a), except we are using only funds that also report FX allocations
c. FX common – as d), except we are using only funds that also report Country allocations
d. FX allocations – the standard Srimurthy-Smalbach-Shen-style twenty-day flow percentage model based on FX Allocations data.
The table below shows performance for ACWI currencies using these four approaches. One-day flow percentage has, in each case, been subset to the period between 1Q15 and 1Q20. This ensures that four are based on full coverage and are tested over precisely the same period.
As you can see from the table below, when you reduce the country allocations dataset to only those funds also reporting FX allocations, the performance of the predictor gets weaker.
Overall, using cross-border equity funds FX allocations data does not yield superior - or even similar - results as our classical predictor driven by country allocations data.
Throwing everything at the problem
Faced with these results, we looked for some new approaches. As noted, the earlier analysis focus on cross-border funds only. What happens when we run a similar analysis, but this time use the full FX allocations database?
The results, as can be seen below is weaker for ACWI and G10 spaces compared to our country allocations-based model. But, significantly, the signal for EM is a solid contrarian one (see table below).
To dig deeper into why and how the full FX allocation dataset generates a strong contrarian signal, we next ran a similar analysis using the following four predictors:
a) All funds reporting FX allocations
b) FX allocations using only those funds common to country
c) As b), but adding in country allocations
d) Full country allocations universe
The results, as can be seen below, clearly show the added value of the FX Allocation database in generating contrarian signals for EM.
Using the whole country allocations database (including the bond funds) also generates a contrarian signal in the same period. Compared to FX Allocations database, however, this signal is weaker, and adding more FX allocation information increases the accuracy of the contrarian signal.
One potential explanation for this contrarian strategy may be that, within EM, currencies paying higher interest rates have been receiving less flows. We would expect the opposite – higher interest rate paying currencies should receive higher flows – and this may be capturing a short-term signal that will revert over time.
One hint of this is that, over time, Q5’s overshoot, and the contrarian signal that drives, is significant in frequencies up to one month but then disappears.
For more insight subscribe here.