Você está na página 1de 3

Systematic Methods for Asset Allocation

Eugene A Durenard
Eric S Hirschberg
Jul-08

Introduction
We present an example of the application of systematic trend-following models to create a
long only asset allocation strategy that outperforms the static mean-variance optimal
allocation.

Our goal is to illustrate the applicability of systematic strategies to asset-liability management


problems, an area where such techniques have not been widely introduced up to now. It is
our view that such an approach presents a clear business advantage by transferring some
know-how and experience from the proprietary trading business to the institutional
allocation side of the business thereby enhancing overall returns for the allocator.

Static Analysis
In the following example we take 4 asset classes, represented by SPX, the 10Y US Yield
benchmark, the CRB Index and the Overnight Repo rate, from 1980 to 2008. We create
daily return series from those.

By the classic mean-variance analysis we come up with a deliberately over-fit optimal


portfolio on SPX, 10Y and CRB for the whole period 1980-2008. This portfolio has no
allocation to cash (Repo) – it is long only and fully invested into the 3 assets.

SPX US10YR CRB ON Repo MV_Portfolio


Annual Return 9.56% 9.22% 1.94% 6.00% 8.79%
Annual Vol 16.08% 12.77% 10.36% 0.33% 9.98%
Absolute Sharpe 0.59 0.72 0.19 18.13 0.88

Optimal Weights 0.47 0.45 0.08 0 1

The above optimal portfolio is based on long-term covariances and mean returns. To a large
extent that choice is most of the time sub-optimal as expected returns are proxied on long-
term average returns and subsequent the realised returns tend to have large deviations from
such means. Also the holding time of such a portfolio is usually shorter than the time needed
for realised returns to be “averaged-out” to their long-term means which requires more than
one business cycle.

Dynamic Approach
To address the shortfalls of the optimal MV allocation process, we take the following
approach. We observe that all the assets concerned exhibit sustained multi-week trends. This
is not an observation based on the 1980-2008 sample but on substantially longer time
period* . We will therefore accept that behaviour as a valid component of the return process
and try to exploit it systematically to vary portfolio weights according to a measure of
momentum. The basic idea coming from any trader’s thinking is: “Don’t Fight the Trend!”.
We will therefore underweight assets that have negative momentum, overweight ones with
positive momentum and put the balance into cash (Repo) so as to always maintain a long-
only non leveraged portfolio.

We first construct 3 Trend Indicators TRD_SPX , TRD_10Y and TRD_CRB by taking


same exponential moving averages of prices and yields: TRD_A = EMA(A, 0.01) for each
asset A. (It makes sense to use yields for Fixed Income and not prices as this is how the
market thinks about bond investing).

For each asset A we then build a discrete momentum indicator MOM_A valued at
{NEGATIVE, NEUTRAL, POSITIVE} in the following manner, with neutrality threshold
X = 1.5%:

For Prices:
If A > TRD_A*(1 + X) then MOM_A = POSITIVE
If A < TRD_A/(1 + X) then MOM_A = NEGATIVE
Else MOM_A = NEUTRAL

For Yields:
If A > TRD_A*(1 + X) then MOM_A = NEGATIVE
If A < TRD_A/(1 + X) then MOM_A = POSITIVE
Else MOM_A = NEUTRAL

We then choose the allocations to our Dynamic Portfolio according to the following matrix,
and where rebalancing is done on the close:

SPX US10YR CRB


POSITIVE 0.3 0.5 0.2
NEUTRAL 0.15 0.25 0.1
NEGATIVE 0 0 0

Any excess is invested into the Overnight Repo. So for example if we have a positive
momentum on SPX but negative on bonds and neutral on CRB then our allocation will be
30% to stocks, 10% to CRB and 60% to overnight, with no allocation to 10Y bonds.

Each day we calculate the proportion of the portfolio thus turned over. We attach a
transaction cost of 15bp of nominal value to each trade. It comes out that the average
turnover per trading day is 3% of the portfolio which corresponds to a drag of about 1.2%
per year.

Accounting for those transaction costs we get the following results:

Average Annual Return 7.61%


Average Annual Volatility 5.91%
Absolute Sharpe 1.29
The following graph presents the compounded NAV’s from the static portfolio and the
dynamic portfolio where we adjusted the dynamic portfolio to the same volatility level as the
static portfolio. Although this may appear as a fudge the goal is to show that a 0.5 difference
in Sharpe ratio per year will yield a sizeable difference in wealth longer term due to the
compounding effect. As our exercise is to maximize our terminal wealth with a given risk
constraint such comparison is appropriate:

Figure 1 Dynamic Allocation (Blue) vs Static MV Optimized Portfolio (Pink)

Systematic Asset Allocation :


Comparison of static MV portfolio with dynamic portfolio (equal volatility)

4500

4000

3500

3000

2500

2000

1500

1000

500

0
Jan-80 Jun-85 Dec-90 Jun-96 Nov-01 May-07

Conclusion

This piece is meant to motivate further research and development in the area of Systematic
ALM and Asset Allocation. Orion has developed a toolbox of different models appropriate
for various timescales to help Pensions, Endowments and other dedicated asset allocators to
achieve better risk adjusted returns on their capital.

* Durenard, in house study Orion Investment Management.

Você também pode gostar