Research Perspectives Archive

September 2006

Quantitative Portfolio Analysis

As I have noted before, our research process is centered on meeting with portfolio managers and their teams. When we select funds, we need to find fund managers that we believe are honest, competent, and have a process we believe will add value and be appropriate. The best way to do this is to meet with the portfolio managers that we invest with directly. Our goal is “turn over more rocks” than the competition. Each week, we typically meet with portfolio managers or hear portfolio presentations on 10-15 different funds. Meeting and communicating with managers directly is critical to how we manage money.

However, at Kobren Insight Management, quantitative portfolio analysis also plays a fundamental role in our fund selection process. Before, during and after the manager meetings, we do a great deal of quantitative work — some of it proprietary.

There are two fundamental methods of conducting this sort of quantitative analysis — and we do both. The most common method involves examining the underlying portfolio holdings of a fund and how they evolve. The other method, called returns-based analysis, involves statistically comparing the returns of a fund versus the returns of various stock indices to estimate a fund’s portfolio composition and how it changes over time. Each of these methods has its own merits and drawbacks.

Portfolio-Holdings Based Analysis
Portfolio holdings-based analysis is simply an examination of the actual holdings of an investment portfolio. By looking at current holdings, one can get a reading on a wide variety of market exposures and risk characteristics, including:

• Asset allocation
  
• Top holdings
  
• Portfolio concentration
  
• Sector exposures
  
• Portfolio valuations
  
• Portfolio growth rates
  
• Distribution of style exposures
  
• Distribution of market cap exposures
  
• Potential capital gains
  
• Expected risk exposure
  

While we can collect key insights just by looking at the current holdings on a snapshot basis, it is also revealing to see how the portfolio evolves over time. For example, how does it behave relative to its benchmarks and peers? Perhaps just as important is how it behaves compared to what we expect based on understanding about the fund from our meetings with management. If there are notable differences, how come? If it is a fund that we own, we will want to pick up the phone and find out why sooner, rather than later.

Generally, the lag time between the date of the portfolio data and when it is made available to be analyzed is measured in months. In the best cases, the gap may be weeks, but in the worst cases, it may be six months or even more than a year depending on the data source. Obviously, precision is lost in this case, and that problem is only heightened for investment portfolios that exhibit higher portfolio turnover by rapidly buying and selling its underlying securities.

Even portfolios with low turnover, but high shareholder cash flows (in or out) can significantly alter their holdings over the course of a few months. For example, we recently saw a low turnover fund that had nearly completely changed its top 10 holdings over the course of a year. Forecasting performance based on analyzing its (old) holdings data — which on the surface would seem reasonable given the fund’s low turnover — would have been off target.

The cleanliness and consistency in reporting of holdings can also pose problems for analysis. It takes time to collect the portfolio holdings, make sure all individual securities are accurately identified, and then catalog them using a consistent classification scheme for industry sectors/etc. This is all easier said than done, and it doesn’t sound that easy in the first place!

A related problem is comparing portfolios with two different portfolio dates. For instance, if one is trying to compare two high turnover growth funds, one with holdings from a month ago, and the other with holdings from several months ago, there is a certain degree of confidence lost in the analysis.

Returns-Based Analysis
Returns-based analysis was introduced in the late 1980s by Nobel Laureate William F. Sharpe. Sharpe felt that this analytical process could determine the composition of an investment portfolio solely based on knowing the performance history of the portfolio. In broad terms, it uses a statistical technique called multiple regression analysis to compare a fund’s returns over a particular period with the returns of a group of indices representing the market classification scheme you want (i.e. capitalization indices, sector indices style indices, etc.) and impute how the fund’s holdings breakdown along those classification lines.

Because this method uses up-to-the-minute return data, and you set a consistent classification scheme for all funds in a category through your choice of comparative indices, this approach addresses the two main problems of actual portfolio holdings analysis. With returns-based analysis, a fund analyst can consistently and efficiently examine a large universe of investment portfolios over the same time period

Not only can returns-based analysis help in identifying market exposures, but it can also calculate comprehensive historical risk characteristic analytics such as historical volatility, correlations, beta, downside risk, and many others.

Yet another attractive feature of returns-based analysis is that it is not limited to monthly or quarterly data points like portfolio-based analysis typically is. At Kobren Insight Management, we perform fund and portfolio analysis using daily returns data – often using data points from only a few days before. Using daily data gives us more data points to stay more current on portfolio movement and to provide more confidence in our analysis.

This all sounds rather slick and complicated, and it is! But it is also quite efficient and effective (though not necessarily inexpensive). However as with holdings-based analysis it is not without its own problems.

For those who remember their college statistics course, the quality of a regression analysis is often measured by a statistic called r-squared. Without going into technicalities, a high r-squared means that an analyst can have confidence in the regression; a low r-squared regression translates into lower confidence in the output.

Certain types of funds don’t necessarily lend themselves to generating a high r-squared at all times, such as highly concentrated funds, long/short funds and other “alternative” funds, as well as funds in less-liquid asset classes. Another weakness is that even with a high r-squared, the output is an estimate of a fund’s market exposures. In other words, the output for a particular fund might indicate that it has 20% in technology, but in reality, it has 15%, so precision is not guaranteed.

But the importance of the identification of the market exposures isn’t necessarily in its point-in-time precision, but its ability to catch movement in those market exposures. If we can catch movement in a fund’s market exposures days after it happens – and potentially months before it shows up in the portfolio-based analysis – that is powerful information. If we see these situations, particularly in the funds we own for client portfolios, we know we need to dig deeper into what is going on and potentially act to adjust the over-all portfolio.

Summary
In sum, both types of portfolio analysis are invaluable to us in monitoring the underlying holdings of funds in client portfolios. Because both forms of analysis have their advantages and their weaknesses, by using them in combination we are better able to stay on top of the ever-changing market exposures and risk characteristics of all our portfolios.

 
If you prefer, Rusty's article is also available in a printable PDF format. The PDF will open in a new window. You will need Adobe Reader to view this document - click here to Download Adobe Reader.