Dynamic Hedging of Currency Risk in Investment

Strategies
Qplum Investments Research
November 2018

Abstract
Often, investors fully hedge their portfolios for currency risk. This can lead
to significant drag in performance for currencies with negative carry. However,
not hedging the foreign currency exposure can lead to significant drawdowns, es-
pecially for conservative investments. In this paper, we consider a conservative,
global tactical asset allocation strategy implemented in US dollar denominated se-
curities for a hypothetical, European investor and highlight the benefits of dynamic
currency hedging over static hedging. Using a parsimonious model for hedge ratio
based on multiple features of merit and an explicit check for maximum allowed
under-hedging, we show that a cost aware, dynamic hedging strategy can reduce
the hedging costs substantially while keeping the portfolio risk within mandate
specifications.

* Gaurav Chakravorty (Qplum) and Ankit Awasthi (Qplum). Qplum is a global

investment management firm, which may or may not apply similar investment techniques
or methods of analysis as described herein. The views expressed here are those of the
authors and not necessarily those of Qplum. We would like to thank Dr Michael Steele
for his valuable comments and research assistance. Please refer to important disclosures
at the end of this document.

Electronic copy available at: https://ssrn.com/abstract=3289292

1 Introduction
Table 1 below shows what we are trying to achieve:

Portfolio FX Risk Cost of FX Hedge Total Risk

Original, Unhedged High None Higher than Investment Mandate
100% Hedged None High Within Investment Mandate
Dynamically Hedged Low Low Within Investment Mandate

Table 1: Our objective is to showcase a hedging strategy that lowers cost of currency hedging while
staying within its mandate

To summarize, we will describe a systematic approach to dynamically hedge currency

risk in a manner that aims to maximize expected returns while still containing the risk
to levels prescribed in the mandate. The work we have laid out in this paper ultimately
shows that adjusting the hedging ratio over time to changing market conditions can
potentially yield significantly better performance than that from a constant hedge ratio.
Table 3 illustrates our findings in this paper. For a European investor, investing in a
US Dollar based strategy, dynamic currency hedging achieves much of the reduction in
risk of full currency hedging while simultaneously avoiding much of the cost of currency
hedging. The remainder of the paper will explain how we accomplish this.
In the following sections, first we highlight the need for currency hedging followed
by a discussion on the determinants of currency hedging costs. Next we present our
methodology for dynamic currency hedging and discuss results.

2 Survey of Related Work

Optimal hedging strategies for the U.S. cattle feeder [NL98] is one of the earliest exam-
ples that shows the effectiveness of hedging risks using futures contracts. It shows that
hedging with futures can be an effective way of isolating the real source of income and
remove sources of uncertainty that are tangential to the business model. In this work,
we have chosen expected returns as the objective function. In Optimal hedging with
higher moments [BM13], the authors have looked at a general class of utility functions,
HARA, and tried to optimize hedging rules to maximize that set of utility functions.
The main contribution of that paper is to point out that if the strategy that is being
hedged has significant positive skew or if the securities we are hedging with have a skewed
distribution of returns, then we should look at higher moments in computing the hedge
ratio. As opposed to traditional OLS hedge ratios, a hedge ratio computed by HARA
utility functions would exhibit lower risks and better hedging in out of sample data.
The methods we have outlined here can be extended to higher moments as well. While
others have highlighted that for global multi-asset portfolio, making currency-hedging
decisions asset by asset rather than for the overall portfolio could be sub-optimal[GR17],
the actual optimal hedging calculation is done under long-term capital market assump-
tions considering strategic portfolios[SL16][GR17]. Our methodology is completely data
driven and does not make any assumptions on long-term cost of currency hedging or the
risk of the currency or asset allocation. And while we show results for a EUR portfolio,
our method is agnostic to base currency and can be used across different base currencies.

3 Need for Currency Hedging

If someone invests his assets in one currency into a strategy that is denominated in
another currency, he must take into consideration that he may lose out on returns after

Electronic copy available at: https://ssrn.com/abstract=3289292

converting the assets back into his original currency. For instance, say an investor with
assets in EUR is invested in a USD denominated strategy. Then, they would be taking
on a EUR/USD currency risk. Even if the strategy returns were flat but the US Dollar
drops 10% against the Euro in that year, the overall net return for the investor would
be down 10%. In this scenario, by hedging currency risk, the investor would have been
able to avert most of this loss attributed to the EUR/USD exchange rate.

Performance Statistics Fully-Hedged Strategy Unhedged Strategy

Annualized returns 4.75% 6.55%
Annualized volatility 4.9% 10.7%
Worst Drawdown 10.4% 25.5%
Table 2: Backtested Return in EUR) and risk profile of a conservative global tactical asset allocation
strategy [See Appendix for details on the strategy] with and without hedging. Please note that these
are backtested gross return numbers and not net of fees. However, the conclusion would not change if
this is measured net of fees

Table 2 shows the returns of two Global Tactical Asset Allocation1 strategies as
measured in EUR. The underlying strategy is same in both cases. The only difference
between the two is that in the “Fully Hedged Strategy”, the USD exposure has been fully
hedged by buying EUR/USD currency futures. The strategy has annualized volatility of
4.9% when fully hedged with EUR/USD currency futures. However, the same strategy
has annualized volatility of 10.7% without hedging, which is much higher than the
mandate of a typical conservative GTAA investment strategy. Moreover, it is important
to note the dramatic increase in the worst drawdown: a 10.4% in the fully-hedged
portfolio versus 25.5% in the unhedged portfolio.

4 Dynamic Currency Hedging

We established that currency hedging is important specially for a low-risk investment
strategy to remain low-risk. However, currency hedging can lead to significant per-
formance deterioration as shown in table 2. The returns go down from 6.55% for the
unhedged strategy to 4.75% for the fully hedged strategy. We propose a strategy for
dynamic hedging that is cost aware. We use a number of previously researched features
to model cost of hedging - these are discussed in the next section. Intuitively, the dy-
namic hedging strategy reduces the hedge ratio when cost of hedging is high and tries
to remain hedged when the cost of hedging is low. In other words, the strategy takes
on currency exposure only when the cost of hedging could be substantially high or the
gains from not hedging could make up for currency movements. In the next section, we
discuss details and implementation of the proposed dynamic hedging strategy.

5 Implementation of Dynamic Hedging Strategy

Implementation of dynamic hedging could be quite different for different currency pairs.
In this section, we will show the back-tested performance of a dynamic hedging strategy
for a European investor with stake in a US Dollar strategy. In future work, we will
expand the discussion to currency pairs which have shown significant skew in their
returns.
Since, our strategy involves reducing the hedge ratio, we will instead come up with
under-hedging in the portfolio. The dynamic hedging fraction is computed as follows
1 See Appendix for details

Electronic copy available at: https://ssrn.com/abstract=3289292

 U HR(t) = M (t) ∗ C(t) (1)

UHR(t) denotes the under-hedging ratio at time t

M(t) denotes the maximum allowed under-hedging as a fraction, M (t) ∈ (0, 1)

C(t) denotes the normalized cost of hedging, C(t) ∈ (0, 1)

Dropping the index t over time in the next set of equations for ease of notation. As
an european investor, investing in USD denominated strategy requires buying USD and
selling EUR.
gta gta
U HReur = Rusd − Rs (2)

gta
Rusd denotes log returns of underlying GTAA portfolio in USD

gta
U HReur denotes log returns of unhedged GTAA portfolio in EUR

Rs denotes log returns of EUR/USD spot

For calculating the returns of the fully hedged strategy, we buy the respective futures
contract to hedge currency exposure2 .
gta gta
F HReur = Rusd − Rs + Rf + m ∗ CB (3)

gta
F HReur denotes log returns of fully hedged GTAA portfolio in EUR

Rf denotes log returns of EUR/USD futures

m denotes the margin required for the futures position as a fraction3

CB denotes the cost of borrowing USD for margin requirements4 .

Finally, the returns of dynamic hedging strategy overlaid on the GTAA portofolio is
computed
gta gta
DHReur = Rusd − Rs + (1 − U HR) ∗ Rf + m ∗ (1 − U HR) ∗ CB (4)

gta
DHReur denotes log returns of dynamic hedging strategy with GTAA portfolio
2 We have chosen to hedge currency exposure using futures since futures are very liquid exchange

traded products and therefore easier to backtest

3 For the purpose of this article, we have taken a margin requirement of 10% for EUR/USD futures

contract traded at CME. To our knowledge, this a fairly conservative estimate

4 Cost of borrowing is approximated using 6-month interest rate.

Electronic copy available at: https://ssrn.com/abstract=3289292

The formulation in equation (1) allows us to work on a model of cost of hedging inde-
pendent of the maximum allowed under-hedging. Moreover, maximum allowed under-
hedging is often a part of mandate specifications of the investment strategy. Let us now
discuss the computation of normalized cost of hedging and maximum allowed under-
hedging.

5.1 Maximum Allowed Under-hedging

As mentioned earlier, if the maximum allowed under-hedging is part of the mandate, our
job here is done. The guidance from investor is often in the form of excess risk that the
dynamic hedging strategy can add. In this case study, we make different assumptions
on investor preferences and propose three sets of experiments
• Static maximum allowed under-hedging - In this case, we assume a risk guidance
that limits the excess volatility added due to dynamic hedging strategy to 10%
of portfolio volatility5 . Then, given the normalized cost of hedging, we solve an
optimization problem to get maximum allowed under-hedging.
The parameters in the problem include the constant maximum allowed under-
hedging M . Also, we need to include the parameters in the formulation for nor-
malized cost of hedging if any (discussed in the next section) - let’s call these θ.
Moreover, let RVmax denote the maximum annual realized volatility allowed post
dynamic hedging6 . The optimization problem is as follows
Pt=T gta
Maximize t=1 DHReur (t), subject to,

gta
RV DHReur ≤ RVmax (5)

gta
RV DHReur denotes annual realized volatility of GTAA portfolio with dynamic
hedging

gta
Note that DHReur (t) is a function of U HR(t) which in turn is a function of M
and C(t; θ)
• Dynamic maximum allowed under-hedging - In this case, we propose a time varying
maximum allowed under-hedging and that it should be function of the expected
volatility of the respective currency (EUR/USD in this case). We approximate the
expected volatility by realized volatility over last 12 months.

M (t) = RV EU R/U SD (t) ∗ k (6)

Note the optimization problem in this effectively remains same given that same as
the first case, we have to learn a constant k.
5 For instance, if the underlying allocation has a volatility of 5%, the addition of dynamic hedging

to the strategy should increase the volatility beyond 5.5% (= 5% + 0.5%)

6 It is important to note that in practice risk metric should be more nuanced - including higher

moments. Here for simplicity, we have just used realized volatility as a metric of risk.

Electronic copy available at: https://ssrn.com/abstract=3289292

Figure 1: The top figure shows the raw short term interest rate spread (difference of EUR 6-month
interest rate and USD 6-month interest rate). The figure in the middle shows the normalized feature
as described in formula (7). The bottom figure shows how the dynamic maximum allowed
under-hedging changes over time.

Electronic copy available at: https://ssrn.com/abstract=3289292

5.2 Normalized Cost of Hedging
In this step our aim is to compute an estimate of future cost of hedging and normalize
it to a value between 0 and 1 which is indicative of the hedging ratio we should employ.
When the under-hedging fraction is 0, we are hedging the portfolio fully. When the
under-hedging fraction is 1, we are under-hedged up to the maximum allowed fraction.

5.2.1 Covered Interest Rate Parity

Returns of buying dollars and hedging currency risk by buying currency futures or
forwards is very close to the interest rate difference between local currency and US
Dollars. Transaction costs and cross currency basis could be significant in case of some
thinly traded emerging market currencies. We can approximate the cost of hedging
primarily by using the difference between the base currency short-term interest rate and
the foreign currency short-term interest-rate. Here base currency is the currency the
investor sees their returns in and foreign currency is the currency that the positions are
being taken in. For example, a European investor buying a security in USD will have
to hedge the USD exposure, and hence the expected cost of hedging is modeled as

rEU R − rU SD

where, rEU R denotes the short term interest rate in EUR and rU SD denotes the short
term interest rate in USD
Figure 1 shows the plot of short term interest rate spread over time. As seen in
the plot the spread tends to be persistent over time. Therefore, besides being a good
estimate of current cost of hedging, short term interest rate spread is also a good estimate
of future cost of hedging. The rationale for this stems from the covered interest rate
parity (CIP) arbitrage trades that would become possible if this relationship did not
hold [Coc70]. This is essentially a risk-less trade that banks could participate in with
their balance sheets [Lev17]. Notable exceptions to covered interest rate parity, either
in tranquil markets[VSM16] or in periods of stress in financial markets, are well worth
understanding. Figure 2 shows the deviations from covered interest rate parity across
time. The central takeaway for our intents and purposes is that while deviations from
CIP happen, it is usually not high for a protracted amount of time, and it is often due
to many wide-spread factors that make prediction difficult[DTV].
An intuitive formulation for normalized cost of hedging is rather simple to come
up for this feature. Lower the interest rate spread, higher is the cost of hedging and
vice versa. Moreover, if the cost of hedging is negative, then remain fully hedged. We
propose the following formulation

C(t) = max(0, min(1, −S(t)/Snorm )) (7)

Here, S(t) is the short term interest spread at time t plotted in figure 1 and Snorm is
5th percentile7 of S(t) over all the available data. Note that this formulation does not
have parameters and the cost of hedging is completely specified and optimization in (5)
is used to obtain just the maximum allowed under-hedging for this case. The formula
(7) suggests that the portfolio should be fully hedged if the cost of hedging is negative.
Moreover, for very high values ( more than Snorm ) of cost of hedging, the portfolio
should be maximally unhedged. Figure 1 shows the value of the normalized feature over
time.
7 This is around -2.3%

Electronic copy available at: https://ssrn.com/abstract=3289292

 Figure 2: The figure shows year-wise deviations from covered interest rate parity. Orange bars
denote the EUR/USD futures returns as predicted by covered interest rate parity, which is simply,
Interest Rate Differential + EUR/USD spot returns.

The above formulation is quite simple and intuitive and demonstrates good results,
but has the following issues:
• The calculation of Snorm has look-ahead bias as it uses all the available data for
estimation.
• The formulation is handcrafted and it would be difficult to generalize this for
features other than short term interest rate spread.
• Due to under-hedging we are exposed to currency risk and there could be other
features which might be helpful for predicting currency movements or cost of
hedging.
Therefore, we need a method for coming with normalized features that is generic and
can easily incorporate multiple features. Moreover, the normalization should be such
that it does not have any look-ahead bias.

5.2.2 Combining Multiple Features

In order to combine multiple features, the first step is to normalize each feature so that
they can be easily combined. Percentile normalization is a generic normalization tech-
nique which gives percentage rank of each new data point with respect to all historical
data points. It is important to note that the normalization is done only on data observed
as of that point in time to avoid any look-ahead bias. Figure 3 shows the plot of feature
after percentile normalization.
One we have normalized the feature values, we can model the cost of hedging as a
linear combination of these features as follows

C(θ) = N F · θ (8)

Electronic copy available at: https://ssrn.com/abstract=3289292

Here, θ is the vector of weights for each normalized feature, N F is the matrix of features
where rows are indexed over time and columns are different feature and C(θ) is the
series of normalized cost of hedging indexed over time. Also, note that in this case, the
optimization in (5) now actually has to find both the maximum under-hedging allowed
and θ

Figure 3: Plot of different normalized features over time. The features based on yield difference are
more persistent over time. The initial flat period for the equity returns spread is the warm-up period

Apart from short term interest rate difference we considered the following features for
predicting currency returns or cost of hedging.

• Long Term Interest Rate Difference : These rate are typically proxies for fixed
income security yields for their corresponding countries. Thus, higher yield rates
tent to signal stronger investor confidence in the economies of those corresponding
countries.
• Difference in returns of stock indices : Similar to fixed income securities, higher
returns of respective stock indices tends to support the corresponding currency as
shown in figure .
• Momentum : Momentum in currency pairs is quite well researched and strategies
that capitalize on momentum actually tend to outperform carry-trades in cur-
rency markets [MS11]. Combination of momentum with other factors have shown
promise in fx markets [Ser10]. Another benefit of using momentum is that it acts
as a of risk management for the currency exposure due to under-hedging. We
elaborate on this further when we discuss the performance of each feature.
Figure 3 shows the plots of normalized features for all the indicators. The hyper-
parameters used for the above features were not optimized for performance - we just
experimented with one set of commonly used values. For long term interest rate differ-

Electronic copy available at: https://ssrn.com/abstract=3289292

 Figure 4: Plot of equal weighted combination of normalized features over time.

ence we used 10 year yields. For stocks returns, we used a lookback of 2 years and for
momentum we used a exponentially trend over last 200 days.

6 Results
6.1 The lure of dynamic hedging
In this section, we aim to show that given any static hedging scheme, we can come up
with a dynamic hedging strategy which improves upon the static hedging scheme while
having equal risk. Using a sigmoid function with just two parameters and the short
term interest rate spread feature, we come up with the cost of hedging. Here we define
a function for under-hedging ratio U HR(t; a, b), as follows:

U HR(t; a, b) = σ(a ∗ S(t) + b) (9)

Recall σ(x) is the sigmoid function, given by

σ(x) = 1/(1 + e−x )

The parameters a and b of U HR(t; , a, b) are learned by maximizing returns under
a given risk constraint. Also, note that the use of sigmoid ensures that our under-
hedging ratio is with in the range (0, 1). So essentially, U HR(t; a, b) will be close to 1 if
hedging cost is high, and T (x) will be close to 0 if hedging cost is low. Moreover, in this
formulation, there is no limit on the amount of under-hedging, so it is possible that the
portfolio remains unhedged completely at times. Next, in order to come up with the risk
constraint, we use a static under-hedging scheme and use the risk of that pnl series as the
upper risk constraint. The optimization is exactly same as that used in (5) except that
here we trying to learn the parameters a and b. We learn for a and b for different static

10

Electronic copy available at: https://ssrn.com/abstract=3289292

 Figure 5: This bar graph encapsulates all the returns (in EUR) numbers we have showcased up-to
this point. On the x-axis is the constant hedging percentage and on the y-axis is annualized returns.
The dynamic hedging for each corresponding static one is done at the same risk level as exhibited by
the constant hedge. In other words, these returns were achieved at equal levels of annualized volatility
for each individual constant hedge. The case of 0% corresponds to the fully hedged case and similarly
the 100% corresponds to the completely un-hedged scenario

hedging schemes and compare performance. Figure 5 shows the performance of dynamic
hedging as formulated above against corresponding static under-hedging schemes.
We don’t suggest using the above formulation for investment purposes because it is
prone to overfitting as we use all the available data for training. The values learnt for
a and b are shown in appendix in table ??. The values indicate sharpe changes in the
hedge ratio as a function of short term interest rate spread which could be disconcerting
to investors. Moreover, we have not put any explicit limit on the maximum allowed
under-hedging here which could be another issue for investors intending to use this
formulation.

6.2 Backtest of Dynamic Hedging Strategies

Dynamic Hedging Dynamic-Hedging

Performance Statistics Fully Hedged Unhedged (Static maximum allowed (Dynamic maximum allowed
Under-hedging) Under-hedging)
Annualized Returns 4.75% 6.55% 5.61% 5.57%
Annualized Volatility 4.9% 10.7% 5.4% 5.4%
Worst Drawdown 10.4% 25.5% 10.4% 10.4%

Table 3: Performance of different hedging approaches based on short term interest rate spread.
Dynamic hedging not only outperforms the fully hedged strategy by saving on the hedging costs but
keeps the risk contained. The most attractive feature of dynamic hedging strategy is that despite taking
on currency risk, it ensures that the worst drawdown is same as the fully hedged portfolio.

First, let’s discuss results for short term term interest rate based cost of hedging as
formulated in (7) with static and dynamic maximum allowed under-hedging as described
in (5) and (6). Figure 1 shows the plots raw feature, normalized feature and dynamic
maximum allowed under-hedging all under the assumed risk guidance of 10% excess

11

Electronic copy available at: https://ssrn.com/abstract=3289292

 Short Term Interest
Performance Static Short Term Interest 10 Year Yield 2 Year Equity Equal Weighted
Rate Spread 200 Day Momentum
Statistics Under-hedging Rate Spread Spread Returns Spread Combination
(%tile Normalization)
Annualized Returns 5.17% 5.61% 5.52% 5.82% 5.31% 5.11% 5.45%
Annualized Stdev 5.4% 5.4% 5.4% 5.4% 5.4% 5.4% 5.4%
Maximum Drawdown 8.2% 10.4% 9.0% 9.7% 8.6% 8.2% 8.1%
Returns to Max
0.63 0.54 0.61 0.6 0.62 0.62 0.67
Drawdown
Annualized Excess Returns 0.41% 0.82% 0.77% 1.05% 0.57% 0.37% 0.71%
Annualized Stdev 2.29% 1.8% 2.0% 2.0% 2.1% 2.1% 2.1%
Maximum Drawdown
12.6% 6.7% 6.0% 5.6% 7.2% 9.3% 6.6%
( Excess Return )
Max Underhedging 30% 40% 42% 35% 24% 41% 40%

Table 4: Performance of different dynamic hedging features. 10 year yield difference has the best
performance in terms of returns whereas the equal weighted combination has the best performance in
terms of returns by drawdown. Most of the features are better or comparable to static under-hedging.
Also, note that the max drawdown of excess returns of all dynamic hedging strategies is better than
the static hedging strategy. Excess returns are computed over the fully hedged strategy. The results
are only reported for static maximum allowed under-hedging9 .
.

volatility due to currency exposure. Using the cost formulation in (7), we got 40% as
the maximum allowed under-hedging. Note that the we make sure that these results
assume a risk guidance in order to compute maximum allowed under-hedging. Table 3
and figure 6 show backtested results for dynamic hedging strategy against fully hedged
hedged strategy. The performance improvement over fully hedged strategy is particu-
larly impressive considering the fact that the worst drawdown after adding currency risk
due to under-hedging remains the same as the fully hedged portfolio. One reason for
this could be the positive correlation of equity markets and EUR/USD [SL16], implying
that during bad times of the underlying GTAA strategy 8 , EUR/USD depriciates as well
which boosts the returns of the currency exposure due to under-hedging. Admittedly,
this might not be true for other currencies.

Figure 6: Backtested performance of different hedging strategies. The performance of static

maximum allowed under-hedging limit is quite similar to dynamic maximum allowed under-hedging
which uses currency volatility to improve the max under-hedging limit

The dynamic hedging strategy based on short term interest rate spread is essentially
a carry strategy and hence, it relies on relative stability of currency prices and an adverse
movement can eliminate gains arising from the interest rate differential. Moreover, this
feature is unlikely to change quickly which exposes the strategy to similar levels of
currency exposure over extended periods of time and can lead to significant drawdowns.
8 see Appendix for details

12

Electronic copy available at: https://ssrn.com/abstract=3289292

 Using multiple features, diversifies the risk of currency exposure in the dynamic hedg-
ing strategy and alleviates some of these concerns. Figure 3 shows the plot of normalized
features over time. This is used in conjunction with a maximum allowed under-hedging
to come up with the final dynamic hedge ratio. Table 4 shows the performance of
different features (with percentile normalization) individually and in combination as
formulated in (8). In order for us to have a fair comparison, we have added the perfor-
mance of static hedging strategy. We found that around 24% constant under-hedging
achieves similar risk as the assumed risk guidance. All the features other than the dif-
ference of equity returns did better than static hedging strategy in terms of performance
both in terms of returns and drawdown. It is important to note that these normalized
features were generated with absolutely no look-ahead. We also show the results for
equal weighted feature combination, which has the best returns to drawdown ratio. For
weighted combination of features as described in (8), we found that if not using any
optimization, the optimal results are obtained with all the weight given to the best fea-
ture - 10 year yield difference. Since the equal weighted feature combination has similar
performance and much more robust against any potential overfitting, we did not try to
add regularization.
Another important observation is that the performance of the dynamic hedging strat-
egy based on short term interest rate spread with and without look-ahead is quite similar
reinforcing the value of short term interest rate spread in successfully predicting future
cost of hedging.

Figure 7: Year-Wise excess returns for dynamic hedging strategies based on different features. The
spread of data points for a year shows the diversity in the performance of different features.

While it is remarkable that in the results in table 3 showed no increase in worst

drawdown with the addition of dynamic hedging strategy, the results in table 4 are
even more impressive because the worst drawdown is even lower than the fully hedged
strategy.
Let’s further examine the diversification benefits of using multiple indicators. Figure
7 shows yearly excess returns over fully hedged strategy of different features. As shown
in figure 3, in 2017, short term or long term yield difference based features suggest high
cost of hedging and tend increase the amount under-hedging. This resulted in higher
exposure to EUR/USD during 2017, but EUR/USD appreciated more than 14% during
2017 leading to underperformance of the dynamic hedging strategy. As a contrast, the
other two features do not exhibit this behavior in 2017 as shown in figure 3. As a result,

13

Electronic copy available at: https://ssrn.com/abstract=3289292

these two features show relative outperformance in 2017 compared to yield difference
based features. Moreover, given the sharp move in EUR/USD in 2017, the relative
outperformance of momentum feature is in line with our expectations. This further
establishes the utility of using multiple features in conjunction to leverage overfitting.
Lastly, we want to discuss the concern around increase in transaction costs due
to dynamic hedging. The turnover of the dynamic hedging strategy based on equal
weighted feature combination is around 160% a year excluding any turnover due to
rollover. And the turnover due to just rollover is 400% for the fully hedged strategy.
Therefore, additional costs due to turnover from dynamic hedging strategy are not
material.

7 Conclusion and Future Work

We started by demonstrating the inevitability of using a currency hedge and provided
a walk-through of our dynamic hedging methodology. We have established through
various experiments that dynamic hedging strategy which is cost aware is an extremely
attractive option when it comes to addressing high cost of currency hedging. One key
feature of the dynamic hedging strategy was the parsimonious nature of the formulation.
Essentially, with very few parameters and explicit choices, we were able to design a
dynamic hedging strategy that can potentially bring down the cost of currency hedging
while keeping the risk contained within the specified risk mandate and resulting in lower
drawdowns.
Use of multiple features helped reduce losses in bad periods by adding diversity.
Future work includes adding more indicators such as difference of real yields both for
fixed income and equity, relative difference in purchasing power parity through changes
in inflation.
The features that we discussed here are agnostic to the currency pair under consider-
ation - EUR/USD. But it is likely that the same analysis does not work as well for other
currency pairs. In future work, we’d like to replicate the analysis on other currency pairs
and analyze its performance. Another important line of future work includes exploring
non-linear combination of features which could help reduce drawdowns in bad years but
it comes at a cost of potential chances of overfitting.

References
[BM13] Cerny A. Brooks, C. and J. Miffre. Optimal hedging with higher moments.
EDHEC Business School, April 2013.

[Coc70] J. H. Cochrane. The grumpy economist. January 1970.

[DTV] WENXIN DU, ALEXANDER TEPPER, and ADRIEN VERDELHAN. De-
viations from covered interest rate parity. The Journal of Finance, 73(3):915–
957.
[GR17] Helen Guo and Laura Ryan. Currency-hedging optimization for multi-asset
portfolios. The Journal of Portfolio Management, 44(2):100–113, 2017.
[Lev17] R. Levich. A brief survey of measuring and exploiting deviations from covered
interest parity. May 2017.
[MS11] Sarno L. Schmeling M. Menkhoff, L. and A. Schrimpf. Currency momentum
strategies. November 2011.

14

Electronic copy available at: https://ssrn.com/abstract=3289292

[NL98] M. Noussinov and R. Leuthold. Optimal hedging strategies for the u.s. cattle
feeder. OFOR, April 1998.
[Ser10] A. Serban. Combining mean reversion and momentum trading strategies in
foreign exchange markets,. November 2010.
[SL16] Sapra S. and Schloegl L. Optimal currency hedging: A factor investing ap-
proach. PIMCO Quantitative Research, 2016.
[VSM16] Robert N McCauley Vladyslav Sushko, Claudio Borio and Patrick McGuire.
The failure of covered interest parity: Fx hedging demand and costly balance
sheets. October 2016.

Appendix
Underlying GTAA Portfolio
The investment strategy that was used in this case study is a low-risk long-only global
tactical asset allocation strategy. The benchmark of the strategy is 78% Barclays Global
Aggregate TR EurHdg, 12% Custom Global Equity Index, 5% Bloomberg Commodity
index TR EurHdg, 5% liquidity. Custom Global Equity Index is further specified as
60% MSCI World EurHdg, 5% FTSE 100 Index,, 5% FTSE MIB INDEX, and 30%
Euro Stoxx50. The mandate is to provide some excess returns on top of the benchmark.
The intent is to have low turnover of around 100 to 150%. It is at this point natural
to ask that if the portfolio is denominated is investing in global assets why do the
securities have to be USD denominated and take on the currency risk. Note that the
investment strategy is implemented through low cost ETFs which are not available in
other currencies. USD denominated ETFs are attractive given relatively low expense
ratio, high assets under management and have good liquidity. Their relatively high
intraday trading volumes ensure easier implementation for a tactical allocation. Consult
the graph below for reference:

Figure 8: Showcasing the cumulative returns of the Conservative GTAA portfolio versus those of the
specified benchmark. Notice how they are somewhat correlated but the returns of the portfolio are
higher overall.

15

Electronic copy available at: https://ssrn.com/abstract=3289292

Weights of Sigmoid Based Hedging Ratio
The following table lists the weights learned by optimizing the sigmoid cost as described
in formula (9).

Comparison Portfolio Optimal ’a’ Optimal ’b’

10% Constant Under-hedging -4.2 -11.5
20% Constant Under-hedging -10.6 -24.2
30% Constant Under-hedging -355 -765
40% Constant Under-hedging -4.5 -7.8
50% Constant Under-hedging -2.1 -2.3
60% Constant Under-hedging -2.1 -1.1
70% Constant Under-hedging -5.6 -0.9
80% Constant Under-hedging -30.7 7.5
90% Constant Under-hedging -3200 1077

Table 5: Learned values of ’a’ and ’b’ for given comparison hedging schemes. The comparison constant
hedging schemes serve to provide the risk constraint for learning ’a’ and ’b’ optimizing for performance.
Note that the high values of ’a’ implies sharp changes in the hedge ratio as a function of the raw feature
value. The feature used here is short term interest rate spread.

16

Electronic copy available at: https://ssrn.com/abstract=3289292

Disclosures
The factual information set forth herein has been obtained or derived from sources
believed to be reliable but it is not necessarily all-inclusive and is not guaranteed as to
its accuracy and is not to be regarded as a representation or warranty, express or implied,
as to the information, accuracy or completeness, nor should the attached information
serve as the basis of any investment decision. Past performance is not indicative of
future performance.
This document contains hypothetical performance results. Hypothetical performance
results have many inherent limitations, some of which are described below. No represen-
tation is being made that any account will or is likely to achieve profits or losses similar
to those shown. In fact, there are frequently sharp differences between hypothetical per-
formance results and the actual results subsequently achieved by any particular trading
program.
One of the limitations of hypothetical performance results is that they are generally
prepared with the benefit of hindsight. In addition, hypothetical trading does not involve
financial risk, and no hypothetical trading record can completely account for the impact
of financial risk in actual trading. For example, the ability to withstand losses or adhere
to a particular trading program in spite of trading losses are material points which can
also adversely affect actual trading results. There are numerous other factors related to
the markets in general or to the implementation of any specific trading program which
cannot be fully accounted for in the preparation of hypothetical performance results and
all of which can adversely affect actual trading results.
Investing in futures, derivatives or foreign exchange markets is highly speculative
and involves substantial investment, liquidity and other risks. CTA managed accounts
and hedge funds can be leveraged and their performance results can be volatile. Past
performance of issuers, financial instruments and markets may not be indicative of future
results, and there is no guarantee that targeted performance will be achieved.

17

Electronic copy available at: https://ssrn.com/abstract=3289292