21 Toolkit_Attila Rebak_Feb 2012
0 likes93 views
Quantis Investment Management developed the Q-Select system to balance subjective and objective factors in fund selection. The process first filters funds based on criteria like tracking error and fund size. Managers are then scored on people, process, performance, risk monitoring and more. Scores are translated into an expected alpha using a formula accounting for tracking error, breadth, and other factors. A final alpha forecast combines this quantitative forecast with qualitative assessments and past performance to balance type 1 and type 2 errors. Portfolios are constructed to optimize expected alpha and diversification based on the quantitative analysis.
![pLAYERS www.citywireglobal.com
34
citywirEglobal.com
Selectors tool kit
It may look like alchemy but Quantis Investment
Managements Q-Select system combines
subjective and objective facets of fund selection
to achieve a balanced result, says its creator
and the companys CIO Attila R辿bak
W
hen we launched Quantis Investment
Management, we realised we lacked the
economies of scale to practice stock picking on
a global scale. However, a structured fund selection process
can balance subjective and objective factors whether we
build a Chinese equity fund of funds or a Latin American
one. With this in mind we have set up our own process,
called Q-Select, which has similarities with other multi-
managers together with specialities and styles of its own.
The first filter
The first step in our process is to sift for potential funds
that fit into our fund of funds model. Filters used include
tracking error, fund size, consistency, turnover, manager
tenure and standard deviation. The aim is to reduce the
number of candidates to a manageable group of 20-40.
When we have the list of contenders we try to
understand the investment process and the proposition
of the given vehicle. Initially we go through the RFP stage
and roughly complete our scoring template; these points
can then be grouped under the following headings:
People, Process, Performance, Risk monitoring and
management.
This is a common step for many multi-manager
companies, but it is a very subjective one as well.
We are lucky at Quantis because we talk a common
language; there are only minor differences in scores
depending on who completes the scoring system.
We try to meet the decision makers to fully understand
their process, to know what to expect from the fund in
general and in different market situations in particular,
and to go through every ambiguous point with them.
After the meeting we often compare one anothers
scores for the same fund, to reduce the chance of
incorrect or subjective evaluation, and to have a
consensus on the fine-tuned scoring.
Crude alpha forecast
We realised that just a pure score is hardly enough to
structure our fund of funds and determine the weights
of the individual holdings. We turn to the Waring and
Ramkumar article1
to translate our scores into an
expected alpha, which is coupled with a forecasted
tracking-error and the correlation among the alpha of
the given funds, with which we can make a formal
optimisation for our fund of funds.
留 = estimated equilibrium level of alpha of the given
manager;
IC = information coefficient of Quantis, which measures
the correlation between the predicted quality of the
managers and the actual quality of the manager;
= variability in the information coefficient of the
individual managers. As suggested by the original
paper we use 0.07 for all managers;
z = measures the quality of the manager. We translate
our qualitative score with the help of the standard
normal distribution;
Br = breadth of the individual managers, ie the
number of independent bets the managers take
each year;
= tracking error of the manager;
TC = transfer coefficient, how effectively could the
manager translate his view in practice this very
much depends on the number of benchmark holdings
and the active risk of the manager.
We use the formula proposed by Grinold and Kahn for
long-only managers:
where 粒=(53+N)^0.57 and N is the number of securities
in the benchmark portfolio.
The above mentioned translation helps us to establish
a few important links with which we agree:
Firstly, the skill alone (high score) is not enough to add
value; you should factor in risk (tracking error).
Secondly, we can judge whether the fee-structure is
attractive by taking into account the quality and the risk
of the fund. As we forecast before fee alpha with the
translation it is very easy to incorporate the fee structure
of the fund.
Finally, the link is not linear between higher tracking
error and higher (before-fee) alpha. Depending on the
depth of the universe and the turnover of the fund
manager, a lower transfer coefficient can easily eat away
the value added by a higher tracking error. This can be
a concern especially in emerging market countries and
regional funds, and less so for global developed market
or US funds.
As we want to use the tracking error forecast for the
alpha forecast and for risk management as well, we have
to balance the conflicting objectives of the two functions.
For alpha forecast we need a more stable tracking error
as we do not want to trade the underlying funds just
because the forecasted tracking error shrinks from one
A formula
for success
Mgr Mgr
1-粒(N)
Quantis=IC留 賊 z rIC TCMgr Mgr Mgr Mgr
TC=()[]Mgr
1 (1+ )Mgr 1
1 粒()
Mgr Mgr
1-粒(N)
Quantis=IC留 賊 z rIC TCMgr Mgr Mgr Mgr
TC=()[]Mgr
1 (1+ )Mgr 1
1 粒()](https://image.slidesharecdn.com/f5f1bd7d-1e07-4f1b-957b-2b15e48fa1ea-150513152809-lva1-app6891/85/21-Toolkit_Attila-Rebak_Feb-2012-1-320.jpg)
21 Toolkit_Attila Rebak_Feb 2012
- 1. pLAYERS www.citywireglobal.com 34 citywirEglobal.com Selectors tool kit It may look like alchemy but Quantis Investment Managements Q-Select system combines subjective and objective facets of fund selection to achieve a balanced result, says its creator and the companys CIO Attila R辿bak W hen we launched Quantis Investment Management, we realised we lacked the economies of scale to practice stock picking on a global scale. However, a structured fund selection process can balance subjective and objective factors whether we build a Chinese equity fund of funds or a Latin American one. With this in mind we have set up our own process, called Q-Select, which has similarities with other multi- managers together with specialities and styles of its own. The first filter The first step in our process is to sift for potential funds that fit into our fund of funds model. Filters used include tracking error, fund size, consistency, turnover, manager tenure and standard deviation. The aim is to reduce the number of candidates to a manageable group of 20-40. When we have the list of contenders we try to understand the investment process and the proposition of the given vehicle. Initially we go through the RFP stage and roughly complete our scoring template; these points can then be grouped under the following headings: People, Process, Performance, Risk monitoring and management. This is a common step for many multi-manager companies, but it is a very subjective one as well. We are lucky at Quantis because we talk a common language; there are only minor differences in scores depending on who completes the scoring system. We try to meet the decision makers to fully understand their process, to know what to expect from the fund in general and in different market situations in particular, and to go through every ambiguous point with them. After the meeting we often compare one anothers scores for the same fund, to reduce the chance of incorrect or subjective evaluation, and to have a consensus on the fine-tuned scoring. Crude alpha forecast We realised that just a pure score is hardly enough to structure our fund of funds and determine the weights of the individual holdings. We turn to the Waring and Ramkumar article1 to translate our scores into an expected alpha, which is coupled with a forecasted tracking-error and the correlation among the alpha of the given funds, with which we can make a formal optimisation for our fund of funds. 留 = estimated equilibrium level of alpha of the given manager; IC = information coefficient of Quantis, which measures the correlation between the predicted quality of the managers and the actual quality of the manager; = variability in the information coefficient of the individual managers. As suggested by the original paper we use 0.07 for all managers; z = measures the quality of the manager. We translate our qualitative score with the help of the standard normal distribution; Br = breadth of the individual managers, ie the number of independent bets the managers take each year; = tracking error of the manager; TC = transfer coefficient, how effectively could the manager translate his view in practice this very much depends on the number of benchmark holdings and the active risk of the manager. We use the formula proposed by Grinold and Kahn for long-only managers: where 粒=(53+N)^0.57 and N is the number of securities in the benchmark portfolio. The above mentioned translation helps us to establish a few important links with which we agree: Firstly, the skill alone (high score) is not enough to add value; you should factor in risk (tracking error). Secondly, we can judge whether the fee-structure is attractive by taking into account the quality and the risk of the fund. As we forecast before fee alpha with the translation it is very easy to incorporate the fee structure of the fund. Finally, the link is not linear between higher tracking error and higher (before-fee) alpha. Depending on the depth of the universe and the turnover of the fund manager, a lower transfer coefficient can easily eat away the value added by a higher tracking error. This can be a concern especially in emerging market countries and regional funds, and less so for global developed market or US funds. As we want to use the tracking error forecast for the alpha forecast and for risk management as well, we have to balance the conflicting objectives of the two functions. For alpha forecast we need a more stable tracking error as we do not want to trade the underlying funds just because the forecasted tracking error shrinks from one A formula for success Mgr Mgr 1-粒(N) Quantis=IC留 賊 z rIC TCMgr Mgr Mgr Mgr TC=()[]Mgr 1 (1+ )Mgr 1 1 粒() Mgr Mgr 1-粒(N) Quantis=IC留 賊 z rIC TCMgr Mgr Mgr Mgr TC=()[]Mgr 1 (1+ )Mgr 1 1 粒()
- 2. pLAYERS 35Issue 21: February 2012 Selectors tool kit citywirEglobal.com Got something to say on fund selection? Call Jes炭s Segarra Sobral. jsobral@citywire.co.uk +44 20 7840 2175 month to the next and thus the projected alpha is small compared with other funds. But to be able to effectively manage the risk of our fund, we need a tracking error forecast which effectively takes into account every piece of information and responds promptly to market movements. After evaluating many potential candidates we chose exponential smoothing from weekly returns. Final alpha forecast We understand our limitations as well; hence we use other methods to generate the final alpha forecast for the given funds. The aim with this method is to avoid mistakes known in statistics as Type I and Type II errors. Sometimes the manager cannot add value with a well-thought-out investment process due to deficiencies uncovered with our scoring system; or to the contrary the manager can produce alpha because she/he employs information efficiently in a framework of an average investment process. To achieve this goal (ie balance efficiently Type I and Type II error) we combine a priori information (return forecast based on Waring et al. article) with the posterior information (the return achieved during the past 36 months) in a Bayesian fashion, that gives much more weight to our qualitative assessment. Portfolio construction With the expected alpha and covariance matrix at hand we can make a formal optimisation for our fund of funds. There are several advantages to this method: We can identify funds (portfolio composition) which truly diversify our risk compared to the benchmark. With the forecasted risk and expected alpha we can identify the optimal number of funds in our portfolio. Holding only a few funds represents a big risk for us, while a larger number diversifies the expected alpha. A surprisingly small number of funds (four to six) is enough to achieve optimal diversification in case of single country or regional funds. This is also true for global funds, where the fund selector does not want to bet beyond the fund selection (sector, country, region). This quantitative method usually confirms the qualitative assessment of the underlying funds. For example, our optimisation for global developed market equity yielded four funds as the ideal composition, where two of the underlying managers are true stock pickers, one of them tries to add value mostly with sector bets, while the last ones value-added comes mostly from regional calls. We can use several well-defined constraints explicitly during the optimisation process such as minimum weights for individual holdings and tracking error range. Making it work for you As we do not make any asset allocation calls in our strategy, we do not have to make any adjustment to the aforementioned process, but for selectors who want to control the asset allocation as well, our model can be modified. Assuming that the correlation among the alpha of different managers in different asset classes (or regions, countries) is independent, separating the asset allocation and fund selection steps solves the problem. We have been using Q-Select since the beginning and have had positive feedback. Even though the investment process is well structured I have to emphasise the subjective factors, mainly during the selection and evaluation phase. In the phase of fund selection, it is essential to filter the right universe as the alpha forecast depends on the tracking error, which makes sense only when we compare the funds to the correct benchmark. While in the evaluation phase we also need some controlled subjectivity when we score the individual funds and finalise our qualitative assessment. 1 M. Barton Waring and Sunder R. Ramkumar: Forecasting Fund Manager Alphas: The Impossible Just Takes Longer, Financial Analysts Journal March/April 2008. We understand our limitations as well; hence we use other methods to generate the final alpha forecast for the given funds Attila R辿bak CIO Quantis Investment Management