Notes from the IPARM conference in Hong Kong, part 3

The second day of the Third Annual Investment Performance Analysis and Risk Management Asia 2010 (IPARM) conference at the Kowloon Shangri-La in Hong Kong, China began with a panel discussion on the lessons learned from the global financial crisis and the roadmap for 2010 and beyond. Jean-Marc Sabatier, Head of Risk Management Asia for Amundi, started off the conversation talking about how 2010 is the year of opportunity for risk management, particularly in Asia. He cited two reasons:

1. Human resources. Management finally sees the need, nay, the requirement, for a real risk management team in place and are finally willing to pay and support to keep a highly respected team in place.


2. The balance of power has switched (a little bit). In the past, if the risk manager said “no,” it was acknowledged, but then everyone moved on. Now, risk managers are more easily and readily able to say “no” to portfolio and managers and marketing groups and actually carry some weight.

Jean-Marc also added that the job of a risk manager is no longer only about reporting; the job of a risk manager begins with the risk report. Oliver Bolitho, Managing Director from Goldman Sachs Asset Management, then discussed the concept of regret risk which in Asia is related “to a ‘face’ thing that leads to taking logic off the table." Oliver believes this is one of the bigger issues facing the industry as he has see countless examples of portfolio managers turning off their thinking when faced with an investment decision.

From the audience, the panel was asked who should have the final say on the risk of a portfolio? The portfolio manager? Risk manager? Combination of the two? Someone else?

Oliver jumped in first by describing risk as a culture. He continued by saying that if we tried to codify risk, it will get boxed in and will not be there when we need it most. By way of example, if we codified risk (e.g, you could only buy securities with a certain rating), just think what people would have bought in the past few years. Oliver concluded that he thought the risk manager should have the final say, but it should not be up to a single person.

Also in response to the question, Dr. Lincoln Rathnam, CFA, Global Head of Investment Management for EM Capital Management, mentioned the positive experience he had working for an investment management firm that was a partnership and anyone at the firm could say no. Anecdotally, Lincoln thought that having a corporate structure of a partnership was a prime reason why Brown Brothers Harriman escaped relatively unscathed from the financial crisis; everyone at the firm had a veto and they avoided the toxic assets that others so readily accumulated. But ultimately, Lincoln’s answer was that he thinks there needs to be a balance of power between the portfolio manager and the risk manager, one party always wants to say yes, the other party wants to say no, and a middle ground that must be found.

There was also a brief discussion about the “age factor” of risk managers (also known as the “value of experience” to the older demographic). In Asia, and perhaps globally, many risk management teams are junior, i.e., it is often a junior member of the investment management team and all too often someone who has not gone through many of the historical ups and downs. There were no firm answers on how to address this issue, although Jean-Marc mentioned that Amundi recently announced a new policy in which all Portfolio Managers must spend at least three years serving in a risk management capacity. Lincoln made the analogy to General Electric back in the Jack Welch days when he mandated as part of their executive management program that everyone had to spend some time in internal audit.

Next up was Dr. Stan Uryasev, Editor-in-Chief of the The Journal of Risk, who gave a talk on deviation CVaR (Conditional Value at Risk). While this risk measure has been around for a while, as a co-inventor of the methodology, Stan was able to expand on the methodology both in theory and practice. Of particular note, Stan emphasized that CVaR is most useful for risk management, not risk measurement. I strongly encourage those of you interested in learning more to check out his slides that he has made available on his website here.

After serving on the panel, Lincoln pulled double duty with a presentation on stress testing, although to me, the most interesting part of his talk was when he touched on the topic of crisis. Contrary to most, Lincoln believes that “crises are not rare events. We go from crisis to crisis to crisis. It is our nature.” Not a comment you can ignore coming from a man with over 30 years of investment management experience, and to solidify his point, he made reference to a list of financial panics, scandals, and failures. Far from comprehensive, I am sure, but it did cement his point that the next crisis is never too far away as it goes through crises starting in the 17th century and ominously ends with nothing next to number 210.

The final panel of the day focused on finding a risk model or performance system that is appropriate for your investment process. Dr. Laurence Wormold, Head of Research at SunGard APT, started things off with his three pillars of risk analysis:

1. Risk measures: The simple stuff, tracking error, VaR, etc.


2. Attribution: In Laurence’s words, “turning 1 number into 100.”


3. Stress testing and scenario analysis: In his mind, this is the most often ignored aspect of risk analysis as he firmly believes in building shocked market risk models.

A member of the audience immediately jumped in questioning Laurence’s assertion as every firm that he knew of did some form of stress testing. Lawrence acknowledged this, but added that for most firms, stress testing is a box ticking exercise that is largely ignored throughout the company. The stress testing that most firms do lack imagination and is too simplified (e.g., S&P 500 goes down 20%). For the most part, the stress testing that is in the marketplace today suffers from a herd approach; everyone is testing the exact same thing. Laurence further suggested that the current tests should be anchored in economic plausibility; a firm should start from a historical event and then invite colleagues to take that information and think about other ways to create realistic scenarios.

Overall, IPARM Asia was a well organized conference with a very solid slate of speakers and I was quite happy to hear that the organizers have already announced that the fourth annual conference will take place in Hong Kong again next February.

Subscribe by e-mail to receive new Taking Risk posts as they are published.

I walk slowly, but I never walk backwards. Have you moved forward with your risk awareness?

The internet is a huge and seemingly all-knowledgeable place sometimes. When I went looking for a quote using the word "slowly," I not only found the the above by American President Abraham Lincoln, but also a whole 45 minutes-worth of other interesting reading from a huge breadth of people and eras, including this famous and rather apt line from Charles MacKay:

"Men, it has been well said, think in herds; it will be seen that they go mad in herds, while they only recover their senses slowly, and one by one."

There was a consistency in many of the quotes, including these two, underlining the recognition that learning and acceptance can be a slow process which panic and ignorance can quickly offset, but that it is one that we must all stay committed to.

What topic is it that brings me to this introduction? That it is now one year since we saw the huge downward movements in the markets, I want to summarise and comment on what I have seen in terms of change in portfolio risk management over those last 12 months.

Risk Model Providers
Many of the risk models that people were using took some criticism in October/November last year as their models were considered to be slow to react. This in turn lead to the emergence of shorter horizon models (R-squared, Barra, Northfield, etc.) and practitioners were encouraged not to replace their current models, but to complement them with the additional analysis that could now be generated. This parallel analysis can give confidence in markets where the horizons are shifting.

Modelling Techniques
Stress-testing is another complementary analysis technique that got plenty of air-time (not least in this blog) as people have looked to forecast the impact on portfolios of certain market changing events, both historical (e.g., Internet bubble, Rouble crisis) and modeled (e.g., Oil to $200). There has also been work on the incorporation of "fat-tails" into the forecasting models and how they can directly affect the outcome of tests such as the commonly-used 10 day 99% VaR limit. Monte Carlo techniques for analysis of the whole distribution now give us further measure such as Expected Tail Loss and CVaR.

Attribution
When AUM has fallen by >40% it can come across as a little disingenuous to point out that a portfolio outperformed its benchmark by a few points and attribution was probably not the tool at the forefront of people's minds. The subsequent rally experienced since April has however brought it back to centre stage, and the ability to combine risk attribution with the more customary allocation-based methodologies (e.g., Brinson) for both contrast and measurement continues to win approval.

Third Party Commentary
There has been much conjecture following on from last year regarding what, if any, new regulation will be implemented. The criticism of VaR, or at least the over-reliance on it as a single risk measure, was just one high profile area of discussion. The need to improve the granularity, frequency, and depth of reporting is another. The Tower Group recently published a report on risk analysis budgeting in the industry (a summary of which can be heard here) highlighting the expectation of all participants that an increase is on the cards. They report that they expect budgets of IT spend as a whole to remain fairly flat but that the allocation of those budgets towards the understanding of risk will rise markedly.

Summary
So what have we been seeing with our clients? The drop in the market from September last year brought a reduction in AUM which, not surprisingly, was seen in reduced fees and therefore a reluctance to commit to new risk spending. There has been a large amount of general interest picking up in the new models and new implementation techniques. We have seen some clients embrace new analysis and reporting for their business units combining several of the above points. But most are undecided and unwilling to commit to change, perhaps balancing the recognition that things have to change with demanding investment into an area seen as a necessary evil rather than the necessary toolset that I believe risk analysis to be.

What are you doing? Which direction do you believe improvement in risk awareness will come from? Are you moving forward? In the financial landscape of the moment there seem to be two major players, and as I started with a quote from an American about always looking to progress, I will finish with one from a Chinese man, Confucius:

"It does not matter how slowly you go as long as you do not stop."

To receive future posts by e-mail, subscribe to this blog.

An Alternative Perspective on Risk Management

Much of this blog discusses portfolio risk from the perspective of exposures to market factors and measurements like VaR and tracking error. Though much is debated about their methodology and applicability, most portfolio managers monitor these risk measurements on a regular basis and use them as they see fit. I think most would agree that risk is not something we can encompass into a single number; in risk management, more is better.

Most stock scoring models I encounter are variations of a common construct: combine fundamental factors with momentum factors to generate a multi-factor score. In the continual search for alpha with low risk, a practitioner may want to consider an accounting and corporate governance factor. This often overlooked factor can have the double-impact of raising alpha while lowering portfolio risk.

In this study, I use Audit Integrity’s Accounting and Governance Risk (AGR) Score. In brief, the AGR score measures the accounting and corporate governance profile for North American and Western European stocks: companies with low scores have a higher risk of potentially fraudulent or misleading activity. This type of measurement is traditionally not measured in commercial risk models or is not easily calculated by an analyst.

In a factor test of the Russell 3000, from 12/31/2000 to 6/30/2009, the AGR factor had a statistically significant 12-month information coefficient (IC) of 0.0578. When filtered down to the bottom size quintile, the IC jumps to a significant 0.088. You could look at the AGR data across other cross sections – sectors, valuation bins, etc – and find that the score’s efficacy persists.

The potential for higher alpha portfolios is highlighted in the simulations I run below. Simulations A and B are optimized portfolios* where A uses a short term multi-factor score and B uses the Accounting and Governance Risk Score as the stock scoring measurement.
The results show simulation B having a higher alpha, lower beta, and higher overall IR. Also note the standard deviation of portfolio returns is lower when using the AGR score (i.e., less portfolio risk).

Another practical application of incorporating the AGR score is to see how the AGR can affect a multi-factor stock scoring model. The table below shows the results of portfolios an optimized portfolio using a multi-factor stock score model without (C) and with (D) the AGR incorporated in the score.

Here, despite the higher overall portfolio risk (stdev of portfolio returns), the IR is higher for the multi-factor model that includes the AGR component.

A final example I will walk through is how you can use the AGR as a stop-loss mechanism. The portfolios below were constructed using a trade-rules based simulation†. Portfolio F uses a stop-loss mechanism that sells out of positions that have fallen below acceptable AGR standards.

By using the AGR as a stop-loss mechanism, we are able to turn this negative alpha portfolio (E) into a slightly positive alpha portfolio (F). By keeping a watch on positions with respect to their accounting and risk governance rating, we are able to improve portfolio performance and reduce risk.

The simulations show there are a variety of ways to incorporate this factor into the management of the portfolio. You can gain additional insight by running performance attribution across the AGR groups to see how “very aggressive” companies contribute to your portfolio’s return. In short, adding this alternative risk measurement factor to your analysis can both diversify your stock scoring models and subsequently enhance portfolio returns.

Guest blogger Sammy Choo is Vice President of Quantitative Analytics at FactSet.

*Portfolios A, B, C and D are optimized portfolios with the following constraints:
Asset Min: Max(0,Bench Weight-.5%)
Asset Max: 1.5 x Bench Weight
Sectors: +/- 2% Bench Weight
Expected Returns: Short Term Alpha Score or AGR Score

†Portfolios E and F have the same parameters as A-D, except a rules-based engine is used and stock ranks are Short Term Alphas or AGR Score.

To receive future posts by e-mail, subscribe to this blog.

When Brinson and risk-based performance attribution disagree

While the two approaches to performance attribution explain the same excess return, they are conceptually and mathematically different enough that they will frequently produce inconsistent results. Since the point of analyzing from both perspectives isn’t to instantly validate each other, this kind of discrepancy shouldn’t frustrate you. In this post, I will offer some suggestions on how to proceed when you encounter these types of differences. Our most experienced users jump on these results as an opportunity to gain insight. You can too.

To begin, I should review my terminology. Brinson attribution refers to performance attribution based on active weights. There are different variations, but the effects usually include allocation, security selection, currency, and potentially others. In contrast, risk-based performance attribution decomposes excess return to active risk factor exposures. This is also considered a multi-variate attribution.

Each performance attribution approach has important strengths and weaknesses. In particular, the critical shortcoming of Brinson attribution is that it doesn’t prevent you from selecting a report grouping irrelevant to the portfolio construction process. When mistakenly done, the analysis is not meaningful at best and misleading at worst. Risk-based performance attribution is a good complement to Brinson attribution because it doesn’t suffer from this weakness, and inconsistent results warn you to rethink your report groupings.

Let's review a sample analysis using a fictitious portfolio to help demonstrate the issue (or opportunity). The analysis below compares performance attribution for the last four quarters using both attribution methodologies simultaneously.

It's clear that the two methodologies differ significantly. The Brinson model attributes the excess return almost entirely to security selection. In contrast, the risk-based performance attribution indicates excess return is attributable to both systematic risk exposures and security-specific decisions. Now what?

To gain a deeper understanding, first identify what risk factors were the primary sources of (or detractors from) excess return. This is most concisely studied through the following report which reveals size bias as the largest contributor to systematic excess return.

Now, lets use what we learned from our risk-based performance attribution and change the report grouping from sectors to market cap bins.

This change of grouping clarifies how the portfolio was constructed (whether intentional or not). In retrospect, our mistake was the decision to group the report by sector. In this case, the portfolio construction process did not center on sector selection. The conflicting results in the first report warned us to reconsider our analysis.

Risk-based performance attribution has many positive qualities in and of itself. When combined with Brinson attribution, it is a classic case of the whole being greater than the sum of the parts and that means a more complete analysis that leads to more confident conclusions.

Get new Taking Risk entries delivered to your e-mail as they are posted.

When is Cash Policy Effect appropriate?

For risk decomposition, FactSet and Barra have worked in close cooperation to provide Barra analytical calculations with Barra terminology as an option in Portfolio Analysis. One of the most common questions we get from clients is why there are data and calculation differences between Barra and FactSet when it comes to risk-based performance attribution. One of the most obvious differences is that Barra includes an additional component of the analysis called “Cash Policy Effect” that is absent from risk-based performance attribution in FactSet. This can be a source of confusion, and therefore warrants greater understanding.

What is Cash Policy Effect?
The impact of a cash position is separated out of the analysis and calculated first. The effect is analogous conceptually to cash drag and mathematically to the effect you would expect to see in a Brinson attribution that separated out cash at the top level grouping. After the Cash Policy Effect is computed, the benchmark weights are ratcheted down proportionately by the cash weight so there is no active underweight in equity.

Whether to include or exclude Cash Policy Effect in risk-based performance attribution comes down to a very specific question: If your portfolio has 10% in cash, and a given security has a 4.75% portfolio weight while the benchmark weight is 5% do you consider yourself to be overweight or underweight the security by 25 basis points in your analysis of relative performance?

Of course, the answer depends on the investment process.

Imagine a portfolio constructed quantitatively using an optimizer to achieve the optimal positions and weights. Then, as a purely secondary step, the cash/equity split is determined and that decision may well be more about practical business considerations. Clearly, in this case, the Cash Policy Effect makes perfect sense because you want to back out that cash allocation decision first and then assess the portfolio ex-post as it was truly constructed in the beginning. The intention was absolutely to overweight our selected security by 25 basis points.

But, from our experience, we don’t believe that this is how a typical FactSet client constructs portfolios. FactSet won't assume why the portfolio holds cash. FactSet won't introduce hierarchy into a risk-based performance attribution framework (when the absence of hierarchy is the primary benefit of risk-based performance attribution vs. Brinson/Exposure-based attribution) to account for cash. FactSet won’t adjust benchmark weights inside performance attribution calculations (and thereby re-define active security weights) due to the portfolio's cash position. We believe that most likely you intended to have a 25 basis point underweight.

As I said before, the answer depends on your investment process. FactSet remains committed to offering choice, so inevitably Cash Policy Effect will be another Portfolio Analysis option. We just aren’t willing to make this a development priority until it makes sense for our clients.

So, given this explanation, the logical follow-up question is how significant is the difference? Should I really care? First, the more cash in the portfolio, the larger the Cash Policy Effect. Second, more extreme and disparate security returns lead to larger differences. I created a relatively simple example to help quantify the differences (click to enlarge).


Security J is the example from our base question. Reviewing the results, the most important differences are the directional changes in the Active Weight and Total Effect. If this were your portfolio, would you think of Security J as a positive contributor or a detractor from your relative performance? Active weight is crucial because as you attribute performance to the systematic risk factors, you rely on the active exposure to the factor and that critically depends on the active weight in each security.

In conclusion, the decision to use Cash Policy Effect should depend on your investment process. Its misuse can easily generate highly unintuitive results that reduce confidence in the analysis.

Get new Taking Risk entries delivered to your e-mail as they are posted.

About This Blog

Welcome to FactSet's Taking Risk blog!

FactSet has been a leading provider of financial information and analytic applications for investment professionals around the globe for the past 30 years. We offer instant access to data and analytics to thousands of analysts, portfolio managers, and investment bankers at the world’s premier financial institutions.

For the past nine years, we have leveraged our portfolios analytics capabilities to build a strong risk business that serves the needs of hundreds of our clients. We have partnered with some of the leading risk providers in the world, including Northfield, Barra, APT, Axioma, and R-Squared. Clients rely on us for optimization, backtesting, risk decomposition, and risk based performance attribution. We use Monte Carlo techniques to properly account for the risk associated with derivatives and fat tails and offer stress testing packages to uncover how shocks to different factors will impact the performance of a portfolio.

This blog's authors - the management and senior developers of FactSet’s risk group - have an average of 15 years of industry experience. We will use this blog to generate discussions on current topics in the risk field and welcome your comments and ideas.

Get updates to this blog delivered to your inbox.