• Paul Stafford

Improved FX Proxy Hedging Using Machine Learning

Updated: 3 days ago


Managing FX risk generated in first world countries is readily effected with a deep and liquid set of hedging alternatives. However, investment funds doing business in emerging markets face a double challenge - emerging market currencies (e.g. BRL, MXN, CLP) are typically more volatile, and hedging instruments are often non-existent or prohibitively expensive. When the risk must be managed, the only choice is to use a proxy.

When hedging currencies, it is very difficult to identify a single currency proxy with a useful level of correlation. When the correlation is poor, the portfolio variance and basis risk will be high. Using a "basket" of three to four proxy currencies can improve performance. However, common practice using linear regression) has produced mediocre results and unnecessary basis risk.

Deaglo has taken a more robust and novel approach. New machine learning algorithms such as Ridge and Bayesian Regression overcome many of the limitations of the traditional approach. They are less sensitive to noise, less likely to over-fit and make the most of available data (through train/test splits and k-fold cross-validation). They are computationally intensive but produce excellent results.

The machine learning approach works very well for numerous EM currencies. Fig 1 shows two examples, hedging USDCLP (Chilean peso) and USDCOP (Columbian peso) with four proxies.

fig 1

The results are strong- the correlation between both synthetic NDFs and the target asset is above 0.91. Mean errors are in the few percent range.

Operationally, proxy hedging is the creation of a synthetic NDF (non-deliverable forward). An NDF creates a gain or loss in the base currency which offsets any loss or gain between the base currency and the target asset.

Many different currencies have been tested using this approach, using both Ridge and Bayesian regression. Table 1 shows some highlights of that testing across LATAM, African and eastern European currencies:

Efficacy during market dislocations

When the markets are smoothly functioning, the error is clearly minimal, and basis risk low for many EMs. But what happens when a Lehman Bros happens? The correlations aren't as strong, as each economy reacts differently. However, using data from the period of 2007 to 2012, we found that the mean error is very similar, but the std deviations increased materially, generally by about 50%. This would have the effect of reducing the optimum hedge ratio.

Start hedging what you need to

Proxy hedging is a useful strategy when hedging currencies whose hedging instruments are either unavailable/illiquid, or are prohibitively expensive (forward points). This has prohibited many investments in otherwise lucrative markets.

Deaglo's new EM proxy hedging capabilities can now protect these investments. The results for many EM currencies are excellent - the correlation between the asset and synthetic proxy is high (85% - 95%), and prediction errors are very low (average error 2-3%). This makes the hedging of African, Asian and LATAM currencies practical and affordable.

As always, using a proxy hedge offers no ironclad guarantee of efficacy; there will always be a basis risk. When there is no choice but to use a proxy hedge, our machine-learning enhanced strategy vastly reduces your risk.

Deaglo white logo.png
  • LinkedIn - White Circle

All statements of opinion and/or belief contained in this website and all views expressed represent Deaglo Partners LLC's “Deaglo” own assessment and interpretation of information available to it at the date of this website. The information contained in this website is based on publicly available information only, from third-party sources. Deaglo does not represent that it is accurate, timely, comprehensive or complete and it should not be relied upon as such, nor has it been independently verified. No reliance may be placed for any purposes whatsoever on the information contained in this website. Under no circumstances should the information provided in this website be considered as investment or trading advice, or as a sufficient basis on which to make any investment or trading decisions and is provided to you for information purposes only. Whilst this website has been prepared in good faith, Deaglo and its group undertakings, members and employees from time to time (“Affiliates”) do not make and are not authorized to make any representation, warranty, or undertaking, express or implied, with respect to the information or opinions contained in it and no responsibility or liability is accepted by any of them as to the accuracy, completeness or reasonableness of such information or opinions or any other written or oral information made available to you. Without prejudice to the foregoing and to the fullest extent permitted by law, Deaglo and its Affiliates do not accept any liability whatsoever for any loss however arising, directly or indirectly, from use of this website or otherwise arising in connection therewith. Nothing herein shall not exclude any liability for, or remedy in respect of, fraud or fraudulent misrepresentation. Our professionals may provide oral or written market commentary or trading strategies to our clients. Deaglo will not treat recipients as clients by virtue of their receiving this report. Data may be subject to updates and corrections without notice. No part of this material may be (i) copied or duplicated in any form by any means or (ii) redistributed without the prior written consent of Deaglo.