Original Article

A simple scheme for allocating

capital in a foreign exchange
proprietary trading firm
Received (in revised form): 22nd October 2014

Antony Jackson
is Assistant Professor in Financial Economics in the School of Economics at the University of East Anglia in the United Kingdom.
He holds an MA in Economics from the University of Cambridge, an MSc in Management Science & Operational Research from
Warwick Business School, and a PhD in Economics from the University of Leicester. Jackson has an investment banking
background, having been Vice President in Credit Risk for Credit Suisse First Boston in Tokyo. He holds PRIMIA’s Professional
Risk Manager (PRM) designation, and is a Regular Member of the CFA Institute.

Correspondence: Department of Economics, University of East Anglia, Norwich, NR4 7TJ, UK.
E-mail: [email protected]

ABSTRACT We present a model of capital allocation in a foreign exchange proprietary

trading firm. The owner allocates capital to individual traders, who operate within strict risk
limits. Traders specialize in individual currencies, but are given discretion over their choice of
trading rule. The owner provides the simple formula that determines position sizes – a
formula that does not require estimation of the firm-level covariance matrix. We provide
supporting empirical evidence of excess risk-adjusted returns to the firm-level portfolio, and
we discuss a modification of the model in which the owner dictates the choice of trading rule.
Journal of Asset Management (2015) 16, 2–13. doi:10.1057/jam.2014.40;
published online 18 December 2014

Keywords: foreign exchange trading; technical trading rules; portfolio management

The online version of this article is available Open Access

INTRODUCTION impact organizational constraints may have

We present a model in which a foreign on the allocation problem.
exchange trading firm owner shares capital A full mean-variance optimization exercise
among a group of traders. His objective is to (based on the firm-level covariance matrix)
earn excess risk-adjusted returns to the firm- would likely generate a volatile capital
level portfolio, but under the constraint that his allocation scheme. This is due in part to the
employees trade as individuals. They specialize high sensitivity of the optimal weights to
in individual currencies, concentrating solely changes in the problem set-up brought about
on the exchange rate between their designated by time-varying sample estimates of variances
currency and the US dollar. and covariances, but also due to estimation
We contribute to the literature on error. A literature has developed that deals with
risk-adjusted performance measurement the problem – the shrinkage literature.2 We
(RAPM) in two distinct ways. First, we derive propose that the owner adopts a simple risk
a simple plug-in formula for capital allocation budgeting scheme that is based on the
that may be adapted to different risk conditional volatilities of individual currencies,
measures1 and, second, we highlight the but not on the correlations between them.

© 2015 Macmillan Publishers Ltd. 1470-8272 Journal of Asset Management Vol. 16, 1, 2–13
www.palgrave-journals.com/jam/
 Allocating capital in a foreign exchange proprietary trading firm

We discuss two versions of the model: the In this case, the solution provides an optimal
‘discretionary’ and the ‘automated’ model. In level of leverage. The model follows
the former, each trader is tasked with trading a Campbell and Viceira (2002), modified to
single currency, but is given discretion over his the choice between a foreign currency and a
choice of trading rule. In the latter, owners risk-free domestic asset.
dictate both the risk limits and the trading rule The owner exhibits constant relative risk
choices. In both versions, the owner provides aversion (CRRA), with power utility defined
the formula that determines position sizes. over next-period wealth:
The key distinction between the models
Wt1+-1γ
lies in the way information is processed. UðWt + 1 Þ ¼ ; (1)
The discretionary model is an ‘as-if ’ model, 1-γ
in which optimal rules are chosen with the where W is wealth and γ is the coefficient
benefit of hindsight – accordingly, a higher of relative risk aversion.
statistical threshold is put in place if the results We assume portfolio returns are log-
are to be deemed significant. In the automated normally distributed, which implies that
model, the owner’s choice of rule is adapted to next-period wealth is also log-normally
the information actually available in real time. distributed. In conjunction with CRRA
His choice of rule is based on an ever- preferences, the assumption of log-normally
expanding information set, and can be seen as distributed portfolio returns leads to a closed-
an exercise in statistical learning. form solution for capital allocation to the risky
The remainder of this article is set out as asset, equation (18), which is increasing in
follows. In the section ‘Position-sizing expected log returns and decreasing in the
formula’, we develop the position-sizing variance of log returns. Our specification has the
formula for the case of a single trader. In the convenient property that the percentage of
case of multiple traders, we introduce a capital allocated to the risky asset is independent
simple method of risk budgeting that is based of the owner’s current level of wealth.
on an equitable distribution of capital. We The owner’s objective is to maximize the
assume that risk budgets are binding, and that expected utility of next-period wealth:
variations in position sizes are driven primarily !
Wt1+-1γ
by time variation in conditional volatilities. max E
The section ‘Expectations and conditional 1-γ
volatility’ introduces the owner’s forward-
subject to the budget constraint
looking risk-adjusted performance target –  
the target Sharpe ratio acts as the transmission Wt + 1 ¼ 1 + Rp;t + 1 Wt ; (2)
mechanism from conditional volatilities to where Rp,t+1 is the simple portfolio return.
position sizes. The ‘Results’ section provides We detail the algebraic steps necessary to
supporting empirical evidence of the ability restate this problem equivalently in terms of
of both versions of our model to deliver log(Wt ), rather than Wt . The idea is to modify
statistically and economically significant the problem set-up in order to take advantage
excess returns to the firm-level portfolio. of the log-normality of returns assumption.
The final section concludes, and discusses Observe first that, because (1−γ) is a
concurrent research in which we allow traders
constant, maximizing E½Wt1+-1γ =ð1 - γÞ is the
to use their full risk allocations selectively.
same as maximizing E½Wt1+-1γ : Furthermore,
applying the natural log function results in a
POSITION-SIZING FORMULA monotonic transformation, thus maximizing
We begin by solving the owner’s capital E½Wt1+-1γ  is the same as maximizing
allocation problem for the case of one trader. logðE½Wt1+-1γ Þ:

© 2015 Macmillan Publishers Ltd. 1470-8272 Journal of Asset Management Vol. 16, 1, 2–13 3
 Jackson

Our assumption of log-normally Substituting for log(Wt+1) from

distributed wealth now enables convenient equation (8) in equation (7) yields the
manipulation of the awkward-looking objective function
expression logðE½Wt1+-1γ Þ: We make use of   
the standard result of the log-normal density logðE½Wt + 1 Þ ¼ E log 1 + Rp;t + 1 + E½logðWt Þ
function that, if ln(Wt+1) is normally 1   
+ ð1 - γ ÞV log Rp;t + 1 : ð9Þ
distributed, then 2

Of course, E½logðWt Þ is the same as log
E½Wt + 1  ¼ exp E½logðWt + 1 Þ
(Wt ), because Wt is known at time t.
 Constants do not affect the maximization
1
+ V½logðWt + 1 Þ ; ð3Þ exercise, thus maximizing logðE½Wt + 1 Þ is
2
equivalent to the following maximization
problem:
where E and V are the expectation and
  
variance operators, respectively. Taking max E log 1 + Rp;t + 1
natural logs of both sides of equation (3) yields 1   
+ ð1 - γ ÞV log Rp;t + 1 : ð10Þ
logðE½Wt + 1 Þ ¼ E½logðWt + 1 Þ 2
1 After introducing the lower-case notation
+ V½logðWt + 1 Þ; ð4Þ rp,t+1 ≡ log(1+Rp,t+1), we arrive at the much
2
simpler-looking maximization objective:
from which it follows that   1  
      max E rp;t + 1 + ð1 - γ ÞV rp;t + 1 : (11)
log E Wt1+-1γ ¼ E log Wt1+-1γ 2
1   
+ V log Wt1+-1γ : ð5Þ The maximization problem is now
2 stated in terms of log portfolio returns,
which are a non-linear combination of the
By using standard rules for manipulating individual assets in the portfolio. We follow
expectations, variances and natural logs, the Campbell and Viceira (2002) linear
the expression in equation (5) simplifies to approximation of these returns, adapted to the
special case of foreign exchange. The
ð1 - γ Þ logðE½Wt + 1 Þ ¼ ð1 - γ ÞE½logðWt + 1 Þ
equation of the simple return Rt+1 to the
1
+ ð1 - γ Þ2 V½logðWt + 1 Þ; foreign currency is
2
ð6Þ
 S 
t+1
1 + Rt + 1 ¼ 1 + R0;t + 1
*
; (12)
St
and after dividing equation (6) throughout by
(1−γ), we are left with the useful expression where R*0,t+1 is the foreign risk-free interest
rate and St is the exchange rate expressed as
logðE½Wt + 1 Þ ¼ E½logðWt + 1 Þ domestic currency units per unit of foreign
1 currency. Taking natural logarithms
+ ð1 - γ ÞV½logðWt + 1 Þ: ð7Þ throughout equation (12) gives
2

Applying natural logs to the budget logð1 + Rt + 1 Þ ¼ log 1 + R0;t
*
+ 1 + logðSt + 1 Þ
equation, that is equation (2), yields - logðSt Þ: ð13Þ
 
logðWt + 1 Þ ¼ log 1 + Rp;t + 1 + logðWt Þ: The trader allocates αt of his wealth to the
(8) foreign currency and 1−αt to the domestic

4 © 2015 Macmillan Publishers Ltd. 1470-8272 Journal of Asset Management Vol. 16, 1, 2–13
 Allocating capital in a foreign exchange proprietary trading firm

risk-free asset, giving an equation in terms of approximation for excess log returns can be
the portfolio simple return Rp,t+1 of written as

1 + Rp;t + 1 ¼ 1 + αt Rt + 1 + ð1 - αt ÞR0;t + 1 rp;t + 1 - r0;t + 1 ¼ αt r0;t
*
+1 + - r0;t + 1 + st + 1 - st
 
¼ 1 + R0;t + 1 + αt Rt + 1 - R0;t + 1 ; 1
+ αt ð1 - αt Þσ 2t : ð17Þ
2
where R0,t+1 denotes the rate of interest on The final stage is to substitute equation (17)
the domestic risk-free asset. This equation can and the corresponding variance of portfolio
be rearranged to give log returns, α2t σ2t , into the objective function,
  that is equation (11):
1 + Rp;t + 1 1 + Rt + 1 
¼ 1 + αt - 1 : (14)
1 + R0;t + 1 1 - R0;t + 1 max αt Et r0;t *
+1 - r0;t + 1 + s t+1 - s t

1 1
Substituting for 1+Rt+1 from equation (12) + αt ð1 - αt Þσ 2t + ð1 - γ Þα2t σ 2t :
2 2
in equation (14), and taking natural logarithms
throughout, yields The first-order condition yields an optimal
allocation of wealth to the foreign currency of
rp;t + 1 - r0;t + 1 ¼ logf1 + αt ½expðr0;t
*
+ 1 - r0;t + 1 Et ðst + 1 - st Þ + r0;t
*
+ 1 - r0;t + 1 + σ t =2
2
αt ¼ :
+ st + 1 - st Þ - 1g: ð15Þ γσ 2t
(18)
Here we have introduced the lower-case
notation rp,t+1 ≡ log(1+Rp,t+1), r0,t+1 ≡ The proportion of wealth allocated to
log(1+R0,t+1), r*p,t+1 ≡ log(1+R*0,t+1 ) and the foreign currency is increasing in the
st ≡ log(St ). Equation (15) shows that expected appreciation of the foreign currency
portfolio excess returns are a function of the and increasing in the differential between
log interest rate differential, r0,t+1 − r*0,t+1, the foreign interest rate and the domestic
and the log return on the exchange rate, risk-free rate. The proportion is decreasing in
st+1 − st. Let conditional volatility and decreasing in the
owner’s coefficient of relative risk aversion.

+ 1 - r0;t + 1 ; st + 1 - st
*
f r0;t
n h  io
¼ log 1 + αt exp r0;t Portfolio construction
+ 1 - r0;t + 1 + st + 1 - st - 1 :
*

Of course, the owner’s allocation problem is

ð16Þ
complicated by there being multiple
currencies and by each currency being
The second-order Taylor expansion of the assigned to an individual trader. Although we
+1 − r0,t+1 +
*
function f around the point r0,t do not discuss the principal-agent problem
st + 1 − st = 0 yields in this article, employees in proprietary firms
 commonly have individual contracts that
*
f r0;t +1 - r 0;t + 1 + st+1 - st detail percentage profit splits with the owner,
 who in return offers substantially reduced
 f ð0Þ + f ′ð0Þ r0;t *
+1 - r 0;t + 1 + st+1 - st commissions that reflect the large trading
1  2 volumes the firm places with third parties.
+ f ′′ð0Þ r0;t *
+ 1 - r0;t + 1 + st + 1 - st ; That traders act as individuals – rather than as
2
part of a team – restricts the ability of the
with f ′(0) = αt and f ″(0) = αt(1−αt). After owner to allocate capital in accordance with
replacing (r*0,t+1 − r0,t +1 + st+1 − st)2 with its a full mean-variance optimization scheme.
conditional expectation σ2t , the linear The allocations would likely be too volatile,

© 2015 Macmillan Publishers Ltd. 1470-8272 Journal of Asset Management Vol. 16, 1, 2–13 5
 Jackson

with certain traders being allocated large references in the literature to firms’ use of
shares of capital purely on the basis of pairwise threshold levels of risk-adjusted profitability.
correlations between currencies. We suggest Lyons (2001) provides anecdotal evidence
that the owner is more likely to adopt an that foreign exchange trading firms only
equitable scheme, with departures reflecting allocate capital to those strategies expected to
the discipline of traders in staying within their yield annualized Sharpe ratios in the range
risk limits, or by their ability to choose ‘good’ 0.5–1.0; Grinold and Kahn (2000) and
trading rules. As an abstraction, we adopt the Menkhoff and Taylor (2007) suggest that
purely equitable approach and leave for future 0.5 is a common benchmark used for
research the interesting questions of how to identifying ‘good’ trading rules.
measure and reward individual performance. We propose, uncontroversially, that the
We suggest that the owner achieves an owner expects to earn a risk premium as
equitable allocation simply by multiplying compensation for being exposed to exchange
his coefficient of relative risk aversion by the rate risk. Owners adopt a target Sharpe ratio,
number of traders in the firm. Diversification which when rearranged and augmented by a
benefits to the firm-level portfolio ensure trading rule signal gives an expression for the
that portfolio risk lies within the owner’s expected appreciation of each foreign
original risk tolerance. An equitable allocation currency:
can be thought of as a risk-adjusted naive SRtarget ´ σ t
diversification3 strategy without short-sale Et ðst + 1 - st Þ ¼ It pffiffiffiffiffiffiffiffi : (20)
250
constraints. A justification for this approach
could be to avoid the problem of distinguishing Here SRtarget is an annualized measure of
current profitability due to chance from that the Sharpe ratio, It ∈ {−1, 1} is a binary signal,
of trading skill. Logistically, the owner then and we assume that there are 250 trading days
presents a simple position-sizing formula to in a year. Owners use an exponentially
each trader, with variations in position size weighted moving average (EWMA)
reflecting changes in conditional volatility and estimate of volatility. The advantage of the
the level of the firm’s capital: method – which is well established in the
risk management industry – is that it can be
1 Et ðst + 1 - st Þ + r0;t
*
+ 1 - r0;t + 1 + σ t =2
2
used to produce estimates extremely quickly.
αt ¼ ;
N γσ 2t The EWMA estimator is defined by
(19) σ^2t ¼ ð1 - λÞμ2t - 1 + λ^
σ 2t - 1 ; (21)
where λ is a smoothing parameter and μt−1 is
where N is the number of traders. It now
last period’s return. A smoothing parameter of
remains to discuss how the owner calculates
0.94 is generally regarded as appropriate for
conditional volatility and how traders use
daily observations (Alexander and Sheedy,
trading rules in forming expectations.
2010).
Substituting equations (20) and (21)
EXPECTATIONS AND into the capital allocation equation, that is
CONDITIONAL VOLATILITY equation (19), yields the bottom-line
position-sizing formula:
We now introduce the notion of a ‘target !
Sharpe Ratio’. The idea offers a practical 1 It SRtarget r0;t*
- r + 1
pffiffiffiffiffiffiffiffi + + 1 2
0;t 1
αt ¼ + :
solution to the problem of mapping binary Nγ 250σ^t σ^t 2
signals into expectations, but also offers a
(22)
practical insight into the way traders use
simple rules-of-thumb in their decision- Let us examine the sensitivity of the
making. The method is grounded in several position size to each variable in equation (22).

6 © 2015 Macmillan Publishers Ltd. 1470-8272 Journal of Asset Management Vol. 16, 1, 2–13
 Allocating capital in a foreign exchange proprietary trading firm

First, we use a simple risk-budgeting Momentum

scheme (rather than full mean-variance The momentum indicator signals whether the
optimization). To allocate 1/N of optimal rate of change of the exchange rate has been
position sizes to each currency is to be positive or negative over a historical time
conservative; likely there would be benefits period n ∈ {1, 2, 3, …, 250}:
to diversification at the portfolio level that 8
would allow larger position sizes in each < +1 if St > St - n
currency. Second, position sizes are inversely It ðnÞ ¼ 0 if St ¼ St - n :
:
related to the risk aversion, γ, of the -1 if St < St - n
owner. Now, examining the terms in the
parentheses, a higher target Sharpe ratio
implies higher expected returns, and hence Moving average
larger position sizes. The target Sharpe The moving average indicator offers a slightly
ratio – or the expected market price of risk – more complicated version of the momentum
is the mechanism that maps volatility into rule; it includes all sample points in its
expected returns. On balance, however, calculation:
higher estimates of conditional volatility are X t
St
accompanied by smaller position sizes. Even SMAt ¼ :
though the target Sharpe ratio heuristic t¼t - n
n+1
maps higher volatilities into greater absolute The indicator function for the simple
expected exchange rate movements, smaller moving average rule is
position sizes result, as the risk term 8
dominates. Larger position sizes are taken < +1 if St > SMAt
when trading rule signals act in the same It ðnÞ ¼ 0 if St ¼ SMAt :
:
direction as the interest rate differential – -1 if St < SMAt
the ‘carry trade’ effect.

Trading range break

Trading rules The n-day breakout indicator generates a
In the ‘discretionary’ version of our model, positive signal if today’s close is greater than
traders are free to choose the trading rule that the highest high of the previous n prices:
generates signals in their particular currency. 8
< +1 if St >maxðSt - 1 ; St - 2 ; :::; St - n Þ
In the ‘automated’ version, the owner dictates
It ðnÞ ¼ - 1 if St <minðSt - 1 ; St - 2 ; :::; St - n Þ :
the choice of trading rule. We now describe :
It - 1 otherwise
the choice set of trading rules.
We include four types of rule – designed The indicator is zero until the first
to broadly follow those of Qi and Wu (2006), breakout, and maintains this value until a
who in turn apply the stock index rules of breakout occurs in the opposite direction.
Sullivan et al (1999). With the exception of
the ‘filter’ rule, the variable of interest is the
number of days of sample data. There are four Filter
types of trading rule, each having 250 possible The filter rule focuses on recent highs and
parameter values, giving a total trading rule lows. Consider a falling market. The low is
universe of 1000 rules. The ‘momentum’, reset at subsequent lower lows until an
‘moving average’ and ‘trading range break’ n-per cent rise generates a buy signal. We
rules are trend-following rules, whereas the consider n ∈ {1.000 per cent,1.006 per
‘filter’ is a contrarian rule. Each rule is cent,1.012 per cent, …, 2.494 per cent},
described below. designed to capture 250 rules in a range

© 2015 Macmillan Publishers Ltd. 1470-8272 Journal of Asset Management Vol. 16, 1, 2–13 7
 Jackson

consistent with the previous literature. bootstrap (Politis and Romano, 1994)
Sell signals in a rising market are generated generates pseudo-time series of returns by
similarly. sampling blocks of observations from the
The trading rule signals provide the empirical series, where the size of each block
final piece of information required by the is drawn from a geometric distribution with
position-sizing formula, that is equation (22). mean size q. The size of the block is an
We provide empirical evidence that the increasing function of the dependency
owner is able to allocate capital equitably, evident in the empirical data – we use a
while still generating firm-level excess conservative block size of q = 10, as in
risk-adjusted returns. Our exchange rates are Sullivan et al (1999).
drawn from the Federal Reserve Board’s The following iterative procedure
H.10 series, and interest rates are British (White, 2000) obtains the P-value for the best
Bankers Association 3-month LIBOR rates. model. Starting with the first model, and
They cover the period from 4 January 1999 to B = 1000 bootstrap replicate series, the test
20 January 2012, and comprise six major statistic V 1 is defined as
foreign currencies: the Australian dollar, 1
V 1 ¼ n2 R 1 ;
British pound, Canadian dollar, Euro,
Japanese yen and Swiss franc. We now where R1 denotes the mean excess return of
compare the returns to the ‘discretionary’ the first model and n is the number of returns.
and ‘automated’ versions of our model. For each of the B = 1000 bootstrap replicate
series, one calculates the statistic

* 1 *
RESULTS V 1;i ¼ n2 R1;i - R1 ;
(23)
In the discretionary model, the owner dictates i ¼ 1; ¼ ; 1; 000
the position-sizing formula, but allows traders
discretion in their choice of trading rule. where the superscript ‘*’ identifies simulated
This version offers the owner diversification series. The P-value of the first rule is obtained
*
across methods, as well as diversification across by comparing V 1 to the percentiles of V 1;i .
currencies. The discretionary model raises the One then proceeds to examine the second
question of how traders choose their trading trading rule. Compute
n 1 o
rule. Although this is an interesting question
V 2 ¼ max n2 R2 ; V 1 (24)
in itself, we sidestep the modelling problem
by allowing traders to choose the optimal and
in-sample rule for their particular currency. n 1 o
* * *
Clearly this is not achievable in reality, and V 2;i ¼ max n2 R2;i - R2 ; V 1;i ;
thus any statistical inference drawn from the
i ¼ 1; ¼ ; 1000 ð25Þ
exercise must take into account the so-called
‘data snooping’ problem (Lo and MacKinlay, noting that equation (25) uses the same
1990). replicate series as in equation (23).
The Reality Check (White, 2000) tests One proceeds recursively through the
whether the best rule beats the null hypothesis remaining k models, obtaining
of zero excess profitability. The idea is that the n 1 o
researcher can search aggressively across a V k ¼ max n2 Rk ; V k - 1
wide variety of rules, safe in the knowledge
and
that the distribution of the test statistic under n 1 o
* * *
the null hypothesis adjusts to compensate for V k;i ¼ max n2 R2;i - R2 ; V k - 1;i ;
the increased chance of achieving ‘lucky’
results across many searches. The time-series i ¼ 1; ¼ ; 1000:

8 © 2015 Macmillan Publishers Ltd. 1470-8272 Journal of Asset Management Vol. 16, 1, 2–13
 Allocating capital in a foreign exchange proprietary trading firm

The P-value of the optimal rule is obtained of returns by a sequence of 1s and −1s
*
by comparing V k with the percentiles of V k;i : (corresponding to long and short positions),
Table 1 presents the results for and to then analyse the conditioned time
discretionary traders. Panel A presents, for series of returns. Our restricted version of
comparison, the results for the time series of equation (22) merely changes the sequence of
returns conditioned on the trading rule signals conditioning variables to +/−k, where k is
of the optimal rule. To enable meaningful chosen to generate position sizes that are,
comparisons between absolute levels of excess on average, equal to those generated by the
return, we calibrate the coefficient of relative owner’s formula.
risk aversion, γ, to a restricted version of Panel B presents the results for discretionary
equation (22), in which the conditional traders using the optimal rule in combination
volatility estimate is fixed and the trader with the owner’s position-sizing
ignores the interest rate differential: formula. This is an ‘as-if ’ analysis, where
  we study the situation in which each trader
1 It SRtarget
αt ¼ pffiffiffiffiffiffiffiffi  k; chooses the single in-sample rule that is
γ 250σ~t
optimal for their currency. The improvement
where k is a constant. Now γ merely acts as a in Reality Check P-values – as traders actively
leverage parameter – the same percentage of manage their position sizes – is evident across
capital is invested or borrowed for all signals. all currencies. The economic significance of
Reality Check P-values and Sharpe ratios are the results – the Sharpe ratio – increases
unaffected by the particular value of γ chosen, markedly once the traders actively manage
but the calibrated value of γ = 3.7 ensures their position sizes. However, the results in
that, on average, position sizes are equal with Panel B present an interesting dilemma to the
and without the owner’s position-sizing owner. After adjusting for data snooping bias,
formula. An equivalent exercise – and the half of the traders appear to generate excess
one followed by most studies of technical returns. In practice, the owner may have to
trading rules – is to condition the time series balance the competing claims of the star

Table 1: Table presenting the excess returns, Reality Check P-values and Sharpe ratios of individual traders in the
‘discretionary’ model
Best trading rule Round-trip transaction costs

Excess return (annualized) Reality check (P-value) Sharpe (annualized)

0.00% 0.05% 0.00% 0.05% 0.00% 0.05%

Panel A: Without Position Sizing

Australian dollar 103-day momentum 9.4 9.0 0.224 0.271 0.66 0.63
Canadian dollar 113-day momentum 6.2 5.9 0.252 0.327 0.63 0.60
Swiss franc 22-day breakout 5.4 5.1 0.619 0.687 0.47 0.45
Euro 26-day momentum 9.8 9.0 0.046* 0.077 0.93 0.86
British pound 101-day breakout 6.7 6.7 0.217 0.235 0.68 0.67
Japanese yen 176-day breakout 5.5 5.5 0.514 0.531 0.54 0.54

Panel B: With Position Sizing

Australian dollar 1.7% filter 21.0 19.6 0.009** 0.023* 1.08 1.00
Canadian dollar 113-day momentum 12.0 11.0 0.134 0.226 0.73 0.67
Swiss franc 108-day momentum 8.7 7.9 0.452 0.558 0.52 0.46
Euro 26-day momentum 17.2 15.6 0.021* 0.044* 1.04 0.95
British pound 101-day breakout 15.9 15.3 0.046* 0.069 0.87 0.84
Japanese yen 176-day breakout 15.2 14.7 0.100 0.131 0.73 0.70

All simulations use a target Sharpe ratio of 0.5 and deduct 0.05 per cent proportional round-trip transaction costs, as
in Qi and Wu (2006).
Levels of significance are * = 5 per cent and ** = 1 per cent.

© 2015 Macmillan Publishers Ltd. 1470-8272 Journal of Asset Management Vol. 16, 1, 2–13 9
 Jackson

Table 2: Table presenting excess returns, Reality Check P-values and Sharpe ratios for the firm-level portfolio in
the ‘automated’ model
Best trading rule Excess return Reality check Sharpe
(annualized) (P-value) (annualized)

Panel A: Without Position Sizing

0.00% round-trip costs 108-day momentum 4.4 0.193 0.60
0.05% round-trip costs 108-day momentum 4.0 0.287 0.54

Panel B: With Position Sizing

0.00% round-trip costs 108-day momentum 11.1 0.013* 0.99
0.05% round-trip costs 108-day momentum 10.0 0.024* 0.89

All simulations use a target Sharpe ratio of 0.5 and deduct 0.05% proportional round-trip transaction costs,
as in Qi and Wu (2006).
Levels of significance are * = 5 per cent and ** = 1 per cent.

traders (those delegated with the responsibility be seen that, with the exception of the
of trading the Australian dollar, Euro and Japanese yen, which has been forced into
British pound) with those of the under- losses, individual currency trading remains
performers – traders who still offer value to profitable. In Panel B, one observes consistent
the owner in terms of method and currency improvement in returns once the position-
diversification. This principal/agent problem sizing formula is applied. Interestingly,
is one we are actively researching, but from position sizing turns the Japanese Yen returns
which we abstract in the current article. into profit, suggesting that economic benefit
The corresponding performance of the is being derived from estimating time-varying
firm-level portfolio is described in Table 2. volatility in accordance with the EWMA
The value to the owner in imposing the estimator.
position-sizing formula is striking. Reality In the ‘automated’ model, the owner
Check P-values for constant position sizing assumes control of the trading rule choice.
are 0.193 with transaction costs, and 0.287 We assume the owner instructs every trader
without transaction costs. The Sharpe ratio to use the same trading rule – the best-
of the best rule – the 108-day momentum performing rule on the available historical
rule – is 0.60, reducing to 0.54 when data. At the beginning of the sample – where
transaction costs are properly taken into there is little historical information – the
account. In stark contrast, the returns to trading rule choice is volatile. But eventually
the firm-level portfolio when traders are the owner’s choice converges to the 108-day
charged with using the owner’s position- momentum rule that was found to be the best
sizing formula are both statistically and in-sample rule in Table 2.
economically significant. The 108-day Table 4 presents the results for the firm-
momentum rule is still the optimal in-sample level portfolio in the ‘automated’ model.
rule, but the Reality Check P-value is now Naturally, mean returns are lower than for the
0.024, with a corresponding Sharpe ratio of in-sample ‘discretionary’ model: the best net
0.89. That the Sharpe ratio has also increased returns in Table 2 were 10.0 per cent, whereas
provides evidence that the results have they have decreased to 7.5 per cent in Table 4.
economic, as well as statistical, merit. Nevertheless, the P-value has decreased further
Table 3 examines the impact on individual to 0.021. The reduction in mean returns has
currency traders of being forced to move been more than offset by the elimination of
away from the currency-specific optimal data-snooping bias – the owner only uses
trading rule to the portfolio optimal trading historical information in the ‘automated’
rule, the 108-day momentum strategy. It can model. The qualitative conclusion of the

10 © 2015 Macmillan Publishers Ltd. 1470-8272 Journal of Asset Management Vol. 16, 1, 2–13
 Allocating capital in a foreign exchange proprietary trading firm

Table 3: Table presenting the excess returns and Sharpe ratios of individual traders when all traders use the
108-day momentum rule
108-day momentum

Best trading rule Excess return Sharpe Excess return Sharpe

(annualized) (annualized) (annualized) (annualized)

Panel A: Without Position Sizing

Australian dollar 103-day momentum 9.0 0.63 8.3 0.58
Canadian dollar 113-day momentum 5.9 0.60 4.1 0.42
Swiss franc 22-day breakout 5.1 0.45 4.4 0.38
Euro 26-day momentum 9.0 0.86 4.2 0.40
British pound 101-day breakout 6.7 0.67 4.6 0.46
Japanese yen 176-day breakout 5.5 0.54 −1.7 −0.17

Panel B: With Position Sizing

Australian dollar 1.7% filter 19.6 1.00 17.8 0.87
Canadian dollar 113-day momentum 11.0 0.67 8.9 0.54
Swiss franc 108-day momentum 7.9 0.46 7.9 0.46
Euro 26-day momentum 15.6 0.95 9.3 0.55
British pound 101-day breakout 15.3 0.84 11.4 0.64
Japanese yen 176-day breakout 14.7 0.70 4.6 0.21

All simulations use a target Sharpe ratio of 0.5 and deduct 0.05% proportional round-trip transaction costs, as in Qi
and Wu (2006).

Table 4: Table presenting excess returns and Sharpe (equation (21)) discounts past information is
ratios for the firm-level portfolio in the ‘automated’ determined by the smoothing parameter λ.
model
Excess Time-series Sharpe
Setting λ = 1.00 is equivalent to using a
return bootstrap (annualized) constant volatility estimate, which in
(annualized) (P-value) combination with a target Sharpe ratio,
Panel A: Without Position Sizing implies expectations of a constant risk
0.00% round- 1.5 0.240 0.21 premium. It is evident from the first row of
trip costs
0.05% round- 0.9 0.344 0.12 Table 5 that the worst results follow from an
trip costs assumption of constant volatility. If, however,
Panel B: With Position Sizing the trader uses a value of λ in the region of
0.00% round-
trip costs
8.0 0.020* 0.70 0.94 – as suggested by J.P. Morgan/Reuters
0.05% round- 7.5 0.021* 0.66 (1996) – then the results are robust to
trip costs
variations around this level. The results are
Levels of significance are * = 5 per cent and more forgiving of a trader who errs on the
** = 1 per cent. side of placing greater weight on recent
returns, than of one who errs on the side of
previous section remains – the position-sizing treating volatility as constant. The implication
formula is crucial in generating excess returns is that the owner should be confident using
to the firm-level portfolio. the J.P. Morgan/Reuters (1996) parameter.

Sensitivity analysis CONCLUSION

The owner’s coefficient of relative risk We consider an organizational structure in
aversion and his choice of target Sharpe ratio which a trading firm owner shares capital
are leverage parameters that do not affect among a group of traders, each of whom
the results, other than by scaling the mean trades in isolation from his colleagues. We
returns. Variations in capital allocation are propose an equitable risk-budgeting scheme,
driven by time-varying estimates of volatility. a key strength of which is its computational
The degree to which the EWMA estimator simplicity. The allocation scheme offers

© 2015 Macmillan Publishers Ltd. 1470-8272 Journal of Asset Management Vol. 16, 1, 2–13 11
 Jackson

Table 5: Table demonstrating the importance of the EWMA estimator

EWMA parameter (λ) Excess return (annualized) P-value Sharpe (annualized)

Panel A: ‘Discretionary’ Model

Constant Volatility 1.00 4.8 0.147 0.65
0.98 3.7 0.037* 0.82
0.96 9.7 0.022* 0.87
RiskMetricsTM 0.94 10.0 0.024* 0.89
0.92 10.1 0.026* 0.89
0.90 10.1 0.028* 0.88
0.88 10.0 0.030* 0.86
0.86 9.9 0.035* 0.84

Panel B: ‘Automated’ Model

Constant Volatility 1.00 3.2 0.069 0.41
0.98 6.4 0.030* 0.58
0.96 7.2 0.021* 0.64
RiskMetricsTM 0.94 7.5 0.021* 0.66
0.92 7.6 0.021* 0.66
0.90 7.7 0.021* 0.66
0.88 7.7 0.022* 0.65
0.86 7.7 0.023* 0.64

The smoothing parameter λ = 1.00 is equivalent to a constant volatility assumption; λ = 0.94 is the level
recommended by J.P. Morgan/Reuters (1996). All simulations use a target Sharpe ratio of 0.5 and deduct 0.05%
proportional round-trip transaction costs, as in Qi and Wu (2006).
Levels of significance are * = 5 per cent and ** = 1 per cent.

some stability, with variations in position sizes trade within their limits when in the domain
resulting from changes in conditional of profits, and only at their limits when in the
volatility, rather than from traders’ past domain of losses. Should the owner wish to
performance. We show that this abstraction adopt our equitable structure, his contracts
of the owner’s capital allocation problem will need to be structured accordingly.
generates statistically and economically
significant excess returns to the firm-level
portfolio in both the ‘discretionary’ and NOTES
‘automated’ versions of the model. 1. For a survey of the problem approached from a
We are actively researching extensions to Value-at-Risk perspective, see Aziz and Rosen (2010).
2. See, for example, Ledoit and Wolf (2003, 2004).
the model that reflect the finer organizational 3. Naive diversification is a strategy that allocates an equal
details of actual proprietary trading firms. share of capital to the constituents of a portfolio constructed
One departure from our abstract model is in with a no-sales constraint. See, for example, DeMiguel et al
(2009).
the way owners compensate traders – traders
tend to be self-employed, with formal profit-
sharing contracts. Successful traders see their
capital accounts grow, whereas unsuccessful
REFERENCES
Alexander, C. and Sheedy, E. (eds.) (2010) Advanced value at
traders see their accounts shrink, creating risk models. In: The Professional Risk Managers’ Handbook.
interesting intra-firm capital dynamics. Wilmington, DE: PRMIA Publications, pp. 114–155.
A second extension lies in the way traders Aziz, A. and Rosen, D. (2010) Capital allocation and RAPM.
In: C. Alexander and E. Sheedy (eds.). The Professional Risk
use their risk limits. In this article, we assume Managers’ Handbook. Wilmington, DE: PRMIA
risk limits are binding, as the owner dictates Publications, pp. 12–40.
the position-sizing formula. A relaxation of Campbell, J.Y. and Viceira, L.M. (2002) Strategic Asset
Allocation: Portfolio Choice for Long-Term Investors. New York:
this assumption creates an incentive for traders
Oxford University Press.
exhibiting the traits associated with Prospect DeMiguel, V., Garlappi, L. and Uppal, R. (2009) Optimal
Theory (Kahneman and Tversky, 1979) to versus naive diversification: How inefficient is the 1/N

12 © 2015 Macmillan Publishers Ltd. 1470-8272 Journal of Asset Management Vol. 16, 1, 2–13
 Allocating capital in a foreign exchange proprietary trading firm

portfolio strategy? Review of Financial Studies 22(5): Qi, M. and Wu, Y. (2006) Technical trading-rule profitability,
1915–1953. data snooping, and reality check: Evidence from the foreign
Grinold, R.C. and Kahn, R.N. (2000) Active Portfolio exchange market. Journal of Money, Credit, and Banking
Management: A Quantitative Approach for Producing Superior 38(8): 2135–2158.
Returns and Selecting Superior Money Managers. 2nd edn. Sullivan, R., Timmermann, A. and White, H. (1999) Data-
New York: McGraw-Hill. snooping, technical trading rule performance, and the
Kahneman, D. and Tversky, A. (1979) Prospect theory: An bootstrap. Journal of Finance 54(5): 1647–1691.
analysis of decision under risk. Econometrica 47(2): 263–291. White, H. (2000) A reality check for data snooping.
Ledoit, O. and Wolf, M. (2003) Improved estimation of the Econometrica 68(5): 1097–1126.
covariance matrix of stock returns with an application to
portfolio selection. Journal of Empirical Finance 10(5): 603–621.
Ledoit, O. and Wolf, M. (2004) Honey, I shrunk the sample This work is licensed under a
covariance matrix. Journal of Portfolio Management 30(4): 110–119. Creative Commons Attribution
Lo, A.W. and MacKinlay, A.C. (1990) Data-snooping biases in 3.0 Unported License. The images or other
tests of financial asset pricing models. Review of Financial
Studies 3(3): 431–467.
third party material in this article are included
Lyons, R.K. (2001) The Microstructure Approach to Exchange in the article’s Creative Commons license,
Rates. Cambridge: MIT Press. unless indicated otherwise in the credit line; if
Menkhoff, L. and Taylor, M.P. (2007) The obstinate passion of
foreign exchange professionals: Technical analysis. Journal of
the material is not included under the
Economic Literature 45(4): 936–972. Creative Commons license, users will need to
J.P. Morgan/Reuters (1996) RiskMetrics-Technical Document. obtain permission from the license holder to
4th edn. New York: Morgan Guaranty Trust Company. reproduce the material. To view a copy of this
Politis, D. and Romano, J. (1994) The stationary
bootstrap. Journal of the American Statistical Association license, visit http://creativecommons.org/
89(428): 1303–1313. licenses/by/3.0/

© 2015 Macmillan Publishers Ltd. 1470-8272 Journal of Asset Management Vol. 16, 1, 2–13 13