Вы находитесь на странице: 1из 11

Capital Efficiency in Multi-factor Portfolios

October 17th, 2016










S UMMARY

The debate for the best way to build a multi-factor portfolio mixed or integrated rages on.

Last week we explored whether the argument held that integrated portfolios are more capital efficient than mixed
portfolios in realized return data for several multi-factor ETFs.

This week we explore whether integrated portfolios are more capital efficient than mixed portfolios in theory. We
find that for some broad assumptions, they definitively are.

We find that for specific implementations, mixed portfolios can be more efficient, but it requires a higher degree
of concentration in security selection.

Newfound Research LLC | 425 Boylston Street, 3rd Floor | Boston, MA 02116 | p: 617-531-9773 | w: thinknewfound.com
Case #5067949

October 17th, 2016

In last weeks commentary, we explored the argument as to whether integrated multi-factor portfolios are more capital
efficient than mixed multi-factor portfolios.
As a brief reminder, mixed multi-factor portfolios are those that are built by combining individual factor portfolios.
Integrated multi-factor portfolios are those that are built by selecting stocks with the highest combined factor exposures.
One of the arguments for the integrated approach is that it is more capital efficient because it allows the same dollar to
be invested across multiple factors at once, creating some implicit leverage.
In last weeks commentary, we explored the factor loadings on several mixed and integrated multi-factor portfolios and
found that there was not significant evidence that the integrated approach introduced meaningful capital efficiency.
There were several confounding variables in this analysis, however. Namely, there was limited realized performance
history and the ETFs we evaluated not only include different factors, but also define those factors in different ways.
This week, we wanted to dive into the theoretical side of the argument. Given the exact same factors, would we expect
an integrated portfolio to be more capital efficient than a mixed one?
To simplify our lives a bit, we use the following assumptions:

We are working with normalized factor loadings for each stock. This allows us to work with z-scores (X ~ N(0,1))
instead of arbitrary factor loading levels. This is not too broad an assumption, as most multi-factor portfolios
begin their selection process with this transformation anyway.

Stocks held in each sleeve of the mixed portfolio are given equal weight and each sleeve in the mixed portfolio is
given equal weight.

Stocks held by the integrated portfolio are given equal weight.

A brief word of warning: this commentary is filled with probability theory and calculus. If deriving results is not your cup
of tea, feel free to skip to the later sections, where the results are applied.

The Efficiency of a Mixed Portfolio


To measure the efficiency of the mixed portfolio, we want to ask two questions:
1. If we have N factors, how much exposure do we expect each sleeve to create to its own factor?
2. How much unintended exposure do we expect from factor relationships between sleeves?
To answer the first question, we have to formulate it mathematically. What were really asking is, what is the expected
z-score given that we pick the top alpha% (e.g. 25%) of stocks with the highest z-scores in this category?

Newfound Research LLC | 425 Boylston Street, 3rd Floor | Boston, MA 02116 | p: 617-531-9773 | w: thinknewfound.com
Case #5067949

October 17th, 2016

Put mathematically,
[| '() ]
where '() is the inverse normal cumulative density (CDF) function. If you are not familiar with these types of functions,
think of a rule that tells you what the score cutoff was for a given percentile (e.g. if you were in the 80th percentile on the
ACT, your score would have been a 25).
Let us define:
' = '()
Since X is a standard normal random variable (mean = 0, and variance = 1), then we can solve the above expectation as:
1

3
45

= 1

This equation can be simply solved, since, = = . So we end up with:


' =

3
45

where fx(x) is the normal probability density function (pdf). For a normal random variable with zero mean and positive
variance, this integral has the closed-form solution of:
:

' =

1 ' (45

:
;<5

This provides the expected z-score of the factor sleeve. We then have to divide this value by N to account for the fact
that the sleeve will only be 1/Nth of the total portfolio.
What about the unintended exposure that comes from stocks selected for inclusion in other factor sleeves?
We can formulate our question mathematically as:
[| >() ]
Where X is the factor we are evaluating and Y is another factor within the portfolio, which is possibly correlated to X.
Similarly, we will define:
> = >()

Newfound Research LLC | 425 Boylston Street, 3rd Floor | Boston, MA 02116 | p: 617-531-9773 | w: thinknewfound.com
Case #5067949

October 17th, 2016

We end up with a similar integral:


1

= >

4@

The first step is to solve = . We can do this through a clever trick, defining a third variable, which is
independent from Y:
= ',>
Where the correlation:
',> =

(, )
' >

We can then solve:


=
= + ',> =
= = + ',> =
= + ',>
= ',>
So taking this solution and plugging it back into our integral, we end up with:
',>

3
4@

>

Which is just a generalized form of our first solution (since in the first case ',' = 1).
Putting it all together, what we can say is that the expected exposure the mixed portfolio should have to factor X is:
1

','M [N |N 'M ]
NP)

We think the interpretation of this makes a lot of sense. The total expected portfolio exposure to each factor will be
equal to the average exposure across sleeves. The exposure from each sleeve will be based upon how correlated the
factor z-scores are.
So if correlations between factor z-scores are zero across the board, for each factor we end up with just exposure from
the sleeve dedicated to the factor:

Newfound Research LLC | 425 Boylston Street, 3rd Floor | Boston, MA 02116 | p: 617-531-9773 | w: thinknewfound.com
Case #5067949

October 17th, 2016

1
[| ' ]

If correlations between z-scores are all equal to 1, we would end up with:


1

[N |N 'M ]
NP)

Since the term within the sum is equal for each factor, we end up with:
[| ' ]

The Efficiency of an Integrated Portfolio


For the integrated portfolio, there is really only a single question we have to answer: if we pick the alpha% of stocks with
the highest combined z-scores, what is the expected z-score for each individual factor?
Put in math terms,
= ) + + O
[) | R ]
It is worth pointing out that while each X is distributed as a standard normal random variable, Z is distributed as a normal
variable with mean zero and some positive variance that will be based upon the relationship between Xs.
Solving similarly to the mixed section,
1

3
4T

) = R

where,
) = = 'U,R

1
R

We know that
'U,R =

() , )
R

Solving for the covariance,

Newfound Research LLC | 425 Boylston Street, 3rd Floor | Boston, MA 02116 | p: 617-531-9773 | w: thinknewfound.com
Case #5067949

October 17th, 2016

) , = ) )
= )
= ); + + ) O
= ); + + ) , O + ) O
O

'U,'M
NP)

So we end up with:
) = =

O
NP) 'U ,'M
R;

We can also solve for the variance of Z:


O

R; =

'M,'V
NP) WP)

Putting it all together, we end up with:


) R =

O
NP) 'U ,'M
O
O
NP) WP) 'M ,'V

Our English interpretation of this math is: the expected z-score for this factor is equal to how much this factor
contributed to total score variance (the correlation terms) times the expected total score.
If correlations between z-scores are zero, then we end up with:
1
R

This looks similar to our result in the mixed case, except when correlations are zero, the variance of Z will be N-times the
variance of X. Therefore, the expected value will actually be higher.

Integrated: More Efficient at All Points?


So given these results, can we say that an integrated approach is more capital efficient than a mixed approach?
The answer gets a bit muddled because of all the cross-correlation terms, but we can boil it down to the expected zscore for the mixed portfolio for factor X:

Newfound Research LLC | 425 Boylston Street, 3rd Floor | Boston, MA 02116 | p: 617-531-9773 | w: thinknewfound.com
Case #5067949

October 17th, 2016

','M [N |N 'M ]
NP)

Versus the expected z-score for the integrated portfolio for factor X:
O
NP) ','M
O
O
NP) WP) 'M ,'V

What we want to find is whether the second term is larger than the first for all N and all possible cross-correlations
(assuming the correlation matrix is positive semi-definite). By simply replacing with definitions defined above, crossing
out common terms, and using the following identities:
N N 'M = W W 'V ,
R = R '
We find that the integrated approach is more efficient when:
R <
Since the R [0, ] (the former in the case of two factors with a correlation between z-scores of -1 and the latter in the
case of N factors with correlations between z-scores of 1), the only time a mixed portfolio is as efficient as an integrated
portfolio is when correlations between factor z-scores are equal to 1 or there is only 1 factor in the portfolio.

Relative Efficiency Plotted


Given our assumptions, we have now proven that an integrated approach will be more efficient than a mixed approach,
for any number of factors or cross-correlations.
How much of an impact can this added efficiency have? We can plot some examples to find out.
First, lets example the expected z-score per factor, as the number of factors increases, assuming zero correlation
between factors.

Newfound Research LLC | 425 Boylston Street, 3rd Floor | Boston, MA 02116 | p: 617-531-9773 | w: thinknewfound.com
Case #5067949

October 17th, 2016

Expected Z-Score per Factor

Expected Z-Score per Factor with Varying # Factors


(Assumption: Alpha = 25%; Correlation = 0%)
1.4
1.2
1
0.8
0.6
0.4
0.2
0
1

# of Factors
Integrated

Mixed

We can see that when 5 factors are used, the mixed approach is about half as efficient as the integrated approach.
What happens if we vary our correlation assumption? Assuming a portfolio of only 2 factors:

Expected Z-Score per Factor

Expected Z-Score per Factor with Varying Correlation:


(Assumptions: N = 2; Alpha = 25%)
1.4
1.2
1
0.8
0.6
0.4
0.2
0
Correlation Between Factor Z-Scores
Integrated

Mixed

We can see again that an integrated approach is more efficient for all correlation levels.

Newfound Research LLC | 425 Boylston Street, 3rd Floor | Boston, MA 02116 | p: 617-531-9773 | w: thinknewfound.com
Case #5067949

October 17th, 2016

Revenge of the Mixed Portfolio?


Given these results, is there a way for a mixed portfolio to be more efficient than an integrated portfolio with a slight
change to the assumptions?
The answer is yes. Way back in the beginning, we assumed that the factor sleeves in the mixed portfolio would pick
the top alpha% stocks by z-score and the integrated portfolio would pick the top alpha% stocks by the sum-of-zscores.
If we assume 1000 stocks and alpha is 25%, this means that each factor sleeve will hold 250 stocks (with potential
overlap between sleeves) while the integrated sleeve will only hold 250.
So one way a mixed approach can fight back is by being more concentrated. In other words, instead of picking the top
25%, it might pick the top 10% per factor.
Returning to the math in the third section, this assumption no longer holds:
R = R '
So we end up with the mixed portfolio being more efficient than the integrated portfolio when:
1
'

:
45
( ;

4
( T
1
>

R R

:
;<T

Re-arranging, we get that the mixed approach will be more efficient than the integrated approach when:
ln

R R
R;
';
+ ; > + ln (' )

2
2R

Unfortunately, there is not much more reduction we can do from here since the inverse normal CDF, required to
compute '; , has no closed form. However, we can solve this numerically for a variety of correlations. In a simple 2factor example, we get the following ' assuming R = 0.25.

Newfound Research LLC | 425 Boylston Street, 3rd Floor | Boston, MA 02116 | p: 617-531-9773 | w: thinknewfound.com
Case #5067949

October 17th, 2016

Alpha Level Required in Mixed Approach to Create More Efficient Portfolio than
Integrated Approach
0.3
0.25
0.2
0.15
0.1
0.05
0

What we see is that for a simple 2-factor portfolio, the mixed approach would have to be highly concentrated in its
selection within the sleeves to be more efficient. At a 0% correlation between factor z-scores, we would need to select
the top 9.13% of z-scores per sleeve.

Conclusion
In this commentary we have explored whether an integrated approach is more efficient than a mixed approach for
constructing multi-factor portfolios.
Given some general assumptions, we find that the integrated portfolio should always be at least as efficient as a mixed
approach, and in most cases offers significant efficiency benefits based on the expected z-score of each factor in the
portfolio.

Corey Hoffstein & Justin Sibears

Newfound Research LLC | 425 Boylston Street, 3rd Floor | Boston, MA 02116 | p: 617-531-9773 | w: thinknewfound.com
Case #5067949

October 17th, 2016

To read other commentaries or to subscribe to future posts, please visit blog.thinknewfound.com/category/weekly-commentary.


Certain information contained in this presentation constitutes forward-looking statements, which can be identified by the use of forward-looking terminology such as
may, will, should, expect, anticipate, project, estimate, intend, continue, or believe, or the negatives thereof or other variations or comparable
terminology. Due to various risks and uncertainties, actual events or results or the actual performance of an investment managed using any of the investment strategies or
styles described in this document may differ materially from those reflected in such forward-looking statements. The information in this presentation is made available on an
as is, without representation or warranty basis.
There can be no assurance that any investment strategy or style will achieve any level of performance, and investment results may vary substantially from year to year or
even from month to month. An investor could lose all or substantially all of his or her investment. Both the use of a single adviser and the focus on a single investment
strategy could result in the lack of diversification and consequently, higher risk. The information herein is not intended to provide, and should not be relied upon for,
accounting, legal or tax advice or investment recommendations. You should consult your investment adviser, tax, legal, accounting or other advisors about the matters
discussed herein. These materials represent an assessment of the market environment at specific points in time and are intended neither to be a guarantee of future events
nor as a primary basis for investment decisions. Past performance s not indicative of future performance and investments in equity securities do present risk of loss.
Investors should understand that while performance results may show a general rising trend at times, there is no assurance that any such trends will continue. If such trends
are broken, then investors may experience real losses. The information included in this presentation reflects the different assumptions, views and analytical methods of
Newfound as of the date of this presentation. This document contains the opinions of the managers and such opinions are subject to change without notice. This document
has been distributed for informational purposes only and should not be considered as investment advice or a recommendation of any particular security, strategy or
investment product. This document does not reflect the actual performance results of any Newfound investment strategy or index.
No part of this document may be reproduced in any form, or referred to in any other publication, without express written permission from Newfound Research.
Newfound Research LLC, 2016. All rights reserved.

Newfound Research LLC | 425 Boylston Street, 3rd Floor | Boston, MA 02116 | p: 617-531-9773 | w: thinknewfound.com
Case #5067949

Вам также может понравиться