The illusion of stability, part 1

“The real trouble with this world of ours is not that it is an unreasonable world, nor even that it is a reasonable one. The commonest kind of trouble is that it is nearly reasonable, but not quite. Life is not an illogicality; yet it is a trap for logicians. It looks just a little more mathematical and regular than it is; its exactitude is obvious, but its inexactitude is hidden; its wildness lies in wait.” - G.K. Chesterton

We have seen this happen many times. A financial crisis erupts and everybody, including the New York Times, remembers risk management. There appear lengthy expositions of the falsity of assuming normally distributed returns, and everyone loudly wonders why the industry was not warned by risk models. However, as soon as the situation stabilizes – or rather appears to stabilize – risk management is again relegated back to the specialized conferences and, incredible as it may seem after the last twenty years, tracking error is again used to completely describe the risk profile of the portfolio.

This is not a conspiracy; rather it is the effect of what John Cassidy the illusion of stability. This illusion is supported by a few pillars, each of which I intent to discuss in this and the next few posts. The first pillar is the economic theorizing that comes from looking at the economy as a physician looks at the elementary particles or an astronomer at the galaxy. This is the logician’s trap taught in every institution of higher learning.

In January 2009, the Basel Committee on Banking Supervision finally attempted to disconnect the feeding tubes and get out of the matrix when it proclaimed that:

“most risk management models, including stress tests, use historical statistical relationships to assess risk. They assume that risk is driven by a known and constant statistical process. Given a long period of stability, backward-looking historical information indicated benign conditions so that these models did not pick up the possibility of severe shocks nor the build up of vulnerabilities within the system.”

I know I have used the above quote before. Nevertheless, I keep using it because I believe that it not only provides in a capsule form many of the key misconceptions about risk management, but gives a glimpse of possible ways to deal with them. I have written before about the problems with the present paradigm and the ways of correcting for them. Now that I have put my logical cart before my imaginary horse, let me go back and find the horse. In other words, I would like to briefly discuss the sources of the misconception as I see them, mainly in the economic theories of equilibrium. Why did much of the financial industry believe that “risk is driven by a known and constant statistical process”? Surely, this is not an obvious observation; it requires a certain mindset, a view of the financial markets as a kind of galaxy that we can observe with the telescope and count that the resulting calculations will not need to be changed from day to day.

To understand the source of this view we need to look no further than Leon Valras, a brilliant French economist who was the first to attempt to create a mathematical equilibrium model of the economy. His friends later remembered that he was very inspired by the book on physics that he had then recently read. It fascinated him so much that he vociferously proclaimed his intention to create a new science of political economy, one that would governed by calculus equations just like physics. “Equilibrium” is derived from physics, and it was only natural to look for the equilibrium in the economic system given his premise. One key feature of his model had far reaching implications and it affects virtually every area of economic and financial thinking including risk management. This idea in French is called “totonemont” and it essentially defines the process of gradual adjustment by which participants in the economy slowly move it toward its equilibrium. The process roughly is as follows:

1. Sellers and buyers announce the prices at which they are willing to transact.
2. If the prices match, then the equilibrium is reached according to a set of equations written down by Valras.
3. If the demand and supply are mismatched, prices are altered in increments until the balance is reached, a gradual process called “totonemont.”

As we can see, this model appears to simply follow common sense, something we observe in our daily lives. However, we need to ask a question that is extremely relevant today for any finance practitioner: what makes this change gradual? Why would we assume that the process is constant and stable? The basic answer to these questions given by quantitative economics and quantitative finance is the assumption that supply and demand are relatively stable.

It is interesting to know that Valras explicitly applied this model to financial markets, despite the fact that Europe saw a number of financial bubbles in the18th and 19th centuries. If we are talking about wheat or corn, it might be reasonable to suppose that neither the consumers’ desire to consume them nor the producers’ ability to produce them will change very quickly. The first is limited by the physiology of the human beings, and the second by the physiology of planet Earth. But the situation is quite different for financial assets. There is no obvious limitation on the amount of financial assets buyers are willing to purchase other than the supply of liquidity and credit in the economy. And as we recently saw, the supply of financial assets may grow very quickly if the demand is there (the MBS and CDS markets are only two recent examples).

More so, as German economist Gustav Schmaller observed more than a hundred years ago, demand can actually increase with price. By now it should be obvious that the situations of demand increasing with price or decreasing with it are so common in finance that they almost constitute the rule. This problem lies waste to any attempt to view financial markets as stable or constant. As risk practitioners, we should be aware that there have been and will be periods when supply and demand change drastically, making the models calibrated in normal times useless. When demand for financial assets falls with prices, it creates a wave of demand for and subsequent shortage of liquidity, which is really what is hiding behind the frequently mentioned “rise in correlations.”

The fact that we do not have a stable process that easily lends itself to modeling should not deter us from quantifying the risk of our portfolios and our exposure to such extreme situations. Our primary answer to this problem has been the development of the Event Weighted method of stress testing. This method assumes that the situations in which demand falls with price have many similarities across time, and therefore we can look for similar periods of liquidation to estimate how our portfolio will respond to future instances. This assumption has shown its validity in empirical tests and certainly does not require the leap of faith involved in assuming that financial markets are as orderly as the Solar galaxy in its motion. The financial system may not be stable and constant, but the reaction of the participants in times of instability can be modeled to supply a risk manager with valuable input for decision making.

Don't miss the next post in this series. Receive new blogs by e-mail.

Considerations when implementing a risk management system, part 3

Continuing from my previous posts, I will address the next in our series of questions to consider when implementing a new risk management system. This is the final post of the series. As a reminder, I am summarizing some key points you should consider when selecting a risk system.

Here again are the questions I address:

Part One:

• Who are the Stakeholders?
• Why do you need risk?

Part Two:

• What are your options?
• Where do you need to see risk?

Part Three:

• How should you implement your solution?
• When should this take place?

---

How should you implement your solution?

Let's first address this point: Goodness of Fit does not apply to models only (Risk model selection). We have taken the time to identify the stakeholders, our analytical requirements, and what we need from a risk model. Now all we need to do is pick a model. Here are some things to keep in mind during the evaluation process.

• Do you really understand how the model is constructed and what the output tells you?
  As a risk practitioner it is imperative to have a firm grasp of how a particular model is constructed and how that model’s results are to be interpreted. It’s not enough to understand that a model uses pre-specified factors or principal component analysis to estimate risk. The reality is that any risk vendor worth its salt should be able to provide lots of details regarding model construction and interpretation. In the end, if you don’t understand the model how can you effectively communicate the results to your clients or apply them to your investment decision making process?
  
• Is the provider open up about their methodologies?
  In this day and age, if a model provider operates like a black box, I would look elsewhere. Certainly there may be elements of the risk model creation that are considered proprietary by a third party vendor, but there is really no excuse for a vendor limiting a client’s access to construction details. In my mind, the better you understand a risk model at a fundamental level, the better you can use it to understand your portfolio's risk.
  
• If you have questions, do they have answers?
  Documentation and transparency are important, but there will always be questions unique to your firm and risk provider needs to be able to help you answer and these questions. Ultimately behind every risk model there are people, and this means that during the model selection process you need to evaluate the relationship with the people behind the risk model as much as anything else. If you cannot have an open dialog with your risk provider you are not going to get the most out of the system.

Next up: data (unfortunately there is more to risk than risk models). Risk analysis requires an underlying set of data, which is to say, risk measures are only as good as the data they are based on. So we need to think about what data is actually needed to perform any risk analysis. This is one of the most overlooked points in risk analysis. It boils down to a simple question. Do I want to manage data or manage money? So what are the data sets we should be thinking about?

• Risk Model: Why we need this is should be self evident.
  
• Portfolio data, benchmark data, and pricing: As we all know, in order to generate portfolio risk we need portfolio and benchmark weights. That means we need portfolio and benchmark holdings and quantities and pricing to calculate accurate market values and weights. It is important that you are comfortable with the accuracy of the portfolio and benchmark data.
  
• Security-descriptive data: If there are securities (e.g., derivatives, unlisted, futures, trusts, real estate) in your portfolios that are not covered by the risk model(s) you use, you need a way to increase this coverage. One of the most common means is to supply security terms and conditions. Different firms have different levels of access to terms and conditions data. E.g., If you are Plan Sponsor you may not have a good source for this data and may have to rely on your managers to supply it. Where will you get this data if you need it and how will you store it?
  
• Fundamental and economic data: There are three good reasons to think about this type of data:
  1) To give a clear picture of portfolio’s current situation in a way that makes sense to people unfamiliar with risk, it is often helpful to include other data in the analysis to illustrate a point. For example, if you have underexposure to something like “size” or “value,” it may help your cause to include market cap or valuation measures along with your risk analysis to help with the interpretation.
  
  2) If you plan on optimizing, you will likely want to have the ability to incorporate market data into your models to tilt your portfolio(s) towards real world factors that are important to you investment process.
  
  3) If you plan on applying any stress tests, you will undoubtedly need market data to create the scenarios you wish to test (e.g., rising oil prices, decreasing interest rates, changes in trading volume).
  
• History and timeliness: Make sure that you have a good handle on what kind of history you need and will have access to as well as how often the data is updated. If you are concerned with historical ex-ante risk analysis or optimization, you will need historical data for the portfolio, benchmark, and risk model before you can move forward.

Choosing a risk provider is a big decision, but it should not be made in isolation, so consider the following. Risk models are no longer linked exclusively to the model providers; they are now available through a variety of platforms, integrated to varying degrees. Because of this you may have an opportunity to not only solve your risk needs, but to potentially also meet other needs or solve other problems at your firm unrelated to risk. This could mean consolidating services, saving money (or at least spreading the cost), and minimizing redundant processes.

Scalability and flexibility are particularly important because it may mean you can use one system for multiple purposes within a risk framework and potentially beyond. If you belong to a Risk Team, you may only care about risk itself, but many financial professionals wear multiple hats these days and are interested in several things (e.g., portfolio management, risk, performance, marketing). In the past, you may have needed more than one platform to meet all of these needs. Now if you can find a platform that is both scalable and flexible and still meets your core risk needs, there is a good chance that you can consolidate services, which in turn can lead to cost savings and distribution. In this environment every cost is being scrutinized so any service that allows you to get good value for cost is in demand.

Of course related to all of this is the data behind the scenes. Most investment firms would rather stay away from the business of managing data and stick to their core competencies. As such, make sure you understand how the platforms you are considering integrate data, from your own portfolio holdings to benchmark data to third-party data and beyond. You might find a system that does much of what you need but still requires you to plug in lots of different data sources to get the job done.

Kick the tires! You wouldn’t buy a car without looking under the hood and taking it for a test drive. Implementing a risk system can be difficult, so take advantage of trials, set some goals, and at the very least make sure you have satisfactory answers to the following:

• Is it easy to test out a simple situation?
  If you can’t get results for a domestic equity portfolio easily, don’t hold your breath when it comes to large multi-asset class portfolios.
  
• Is the support responsive?
  If you don’t get the help you need during a trial, forget about when you are client.
  
• How is the software?
  If you can’t use the software, you can’t analyze risk.

When should all of this take place?

There is no perfect timeline for selecting and implementing a risk system, but here is a rough guideline of how it often works:

1. Investigate the needs and requirements internally before anything else. Do as much leg work internally as you can before casting your net and looking at providers. Meetings and demos will be much more effective if you have a good grasp of what you think you need.
   
2. Look at the options available in the marketplace. Do some research about the model types you might be interested in. If possible, attend relevant conferences. Contact vendors and ask for information.
   
3. Meet with the providers and have them explain their solutions in the context of your needs. Risk providers have lots of experience; they should be able to do this and this may force you to re-evaluate your questions.
   
4. Narrow the field. Based on meetings, demos, and conversations, you should have some comfort at this point about who you think are legitimate options.
   
5. Request a trial of the top candidate(s). Keep your goals and objectives in mind. Start thinking about implementation. Perhaps part of the trial can be dedicated to moving forward in this regard.
   
6. Don’t lose focus. Circle back to your original requirements. Are you still on target or have you drifted away from your core goals?
   
7. Purchase approval. Depending on a firm’s purchase approval process this step can sometimes significantly delay or impede implementation. Costs should be discussed early so that there is no confusion or surprise at this stage.
   
8. Full implementation. Fully implementing a risk system may take a while, so create a reasonable time line and start simple or with the key portfolios.

Finally, anything worthwhile tends to be difficult, and I believe that implementing a risk system falls into the category of something that is worthwhile. If you have specific questions, please contact me.

Considerations when implementing a risk management system, part 2

Continuing from my previous post, I will address the next in our series of questions to consider when implementing a new risk management system.

Part One:

• Who are the Stakeholders?
• Why do you need risk?

Part Two:

• What are your options?
• Where do you need to see risk?

Coming in Part Three:

• How should you implement your solution?
• When should this take place?

---

What are your options?

Now that we are clear about who we are trying to please and the reasons we need risk in the first place, we can tackle finding the right risk model for the job. To do this we need to be able answer three primary questions:

1. What market(s) do you invest in?
   
   Firms often try to use a single broad model to analyze many smaller markets that they care about. While this may be a more cost effective option then buying many market-specific models and the large model may, in fact, “cover” all of the securities they care about in the smaller markets, these firms are not taking advantage of the research and development performed by the risk vendors to design their models for specific purposes. For example, it seems to be more and more common to use a Global Equity model to analyze equity portfolios that invest only in single countries. While you will certainly be able to calculate some risk numbers, I would argue the value of these numbers is reduced. Most global models are designed to use or capture factors that apply across many diverse markets and therefore are ideal for global investors, while single country models typically use or pick up factors that are unique to a single market. Imagine using a Global model to analyze an Australian equities portfolio. Most global models will likely be dominated by factors that primarily affect a few large countries; is the predicted risk of such a model ideal for this situation?
   
   
2. What asset classes do you care about (equity, fixed income, derivatives, unlisted assets, etc.)?
   
   You can easily extend the rationale from point 1 to multi-asset class models. If you could, would you use an equity model to analyze a REIT portfolio?
   
   
3. What are the right time horizons (Investment Horizon vs Model Horizon)?
   
   What about the often overlooked time horizon? If some of your portfolios are short term by nature, looking at an estimation of risk based on 12-month time horizon would not be particularly useful. What if you have a long-term investment horizon, but you want to understand your short term risk exposures? There are certainly legitimate and very good reasons to use models calibrated for different time horizons; just make sure that you understand the limitations and applications of such models before making a decision about which model is right for you.

In the end, I think we need to keep sight of something we already know, but often push to the background: all risk models are estimates based on some simplifying assumptions. One model is not going to be ideal for all purposes and situations. We need to do our best to align the model assumptions with our view of the world and our reasons for analyzing risk in the first place. If I care about measuring sensitivity to short-term market volatility, then I should not expect a model with a long horizon to provide meaningful insights.

The ultimate goal should be to fit models to our purposes. In many cases this may mean using multiple risk models.

Where do you need to see risk?

At this point, we have identified (from our list of stakeholders) the people/groups that care about risk, but as I have mentioned, the degree to which these stakeholders care can certainly vary significantly across individuals and groups. My goal here is to help you think about communicating what you know in a way that is meaningful to the end consumer of the information.

I have seen numerous investment managers who already have a risk system in place follow a business model where a small team of risk professionals analyze, generate, and communicate risk results for everyone else in the firm. While there is no denying that having a risk team is still a great idea (who better to set up standards, objectively monitor risk, and communicate results then people who specialize in exactly this type of analysis?), there are definitely some good reasons to make this information more accessible throughout a investment firm. For example, technology and software have improved dramatically since risk systems first came into use, which should allow for much broader and more efficient distribution of risk analytics across a firm. It still may fall to a special risk team to monitor and manage risk across portfolios, teams, etc., but there is certainly no reason for others who might be interested in portfolio risk (e.g., PMs, CIOs, Analysts) to have little or no access to the same information on a regular or ad hoc basis. In other words, there is no technological reason for limiting access to risk data to a single person or team.

At the end of the day, there will likely always be a need to have some risk information “pushed” throughout many firms, but the ability for entirely separate, independent groups to dynamically “pull” data throughout an organization should become standard practice as time goes on. If nothing else, it should allow:

• Better integration of risk within the investment process (e.g., if fund managers can monitor their own risk, they should be able manage their portfolios in such a way that they can better justify investment decisions from a risk/return perspective)
• Improved dialog amongst teams (e.g., if a risk team, board, or CIO is able to monitor risk across portfolios, they can ask meaningful questions to PMs before risk becomes a problem)
• More frequent and timely dissemination of risk results internally and externally

The main takeaway from this section is the need to understand the means and options by which you can communicate analytics across the firm when you are working with your risk model vendor. Make sure that you have scalable solution that can grow as your needs grow and change.

I will be wrapping this subject up in the next installment.

Don't miss the next post in this series. Receive new blogs by e-mail.