What follows are some remarks I intend to make on Friday Oct 21 2011 at a panel on global risk organized by GARP and the Federal Reserve Bank of New York.
I have the luxury of not being a regulator, which I think is a very difficult job. I used to be a physicist, and Nature doesn’t care about regulations; she cares about principles. (If you tried to regulate the motions of the planets you would have a very hard time: turn here, Earth, not too fast, spin more slowly, watch out for the moon, etc. Instead, a few principles of Newton’s take care of everything.) Therefore I’m more partial to principles than regulations, and so I’m going to take the luxury of talking about principles of modeling and the principles of capitalism that, if respected, might mitigate the need for so many regulations.
What I say is based on direct experience of building models in physics and also building and using models at Goldman and Salomon on various derivatives desks, and then at Goldman in Firmwide Risk.
The main question I’m going to address this morning is the one I was given: What is and what should be the roles of models in identifying and managing risk? Should this role differ between the private and public sector.
To answer this I need to clarify the taxonomy of ”models”, a word which can cover a multitude of follies and smartnesses.
The Principles of Modeling
There are two types of models: absolute-value models and relative-value models, and these two types have different purposes and different efficacy.
Absolute-value models are models that aim to predict, without reference to anything else, absolutely, what’s going to happen or what something is worth. In physics, Newton’s principles for the motion of matter or Maxwell’s equations for the propagation of light are absolute models; they work without reference to anything else, and they work astonishingly well – that’s how your iPhones work, and that’s how your Blackberry’s don’t. In a book of mine coming out next week, called Models.Behaving.Badly, I say that absolute models are better called theories, because they stand on their own two legs.
In finance and economics, there are very few absolute models or theories. The efficient market model is an absolute model. It’s nice and beautiful and general, but unlike Maxwell’s equations, it just isn’t true. The truth is, absolute value models in finance and economics suck. They can’t tell you reliably whether something is going up or down tomorrow. Don’t trust them.
Relative-value models stand on someone else’s legs. Relative-value models tell you how something you don’t understand behaves provided you first calibrate the model by telling it how something else you do understand behaves. So for example, in physics you can model the nucleus of an atom as a small drop of liquid provide you calibrate the liquid’s properties to what you observe about the nucleus. But the nucleus isn’t a drop of liquid, though, if you pretend it is, temporarily, you can approximately predict things like nuclear fission. It’s useful but not gospel.
In finance, all useful models are relative-value models. The Black-Scholes-Merton model is an immensely useful relative value model: it says that if you make some naļve but plausible assumptions, then you can create an option out of stock and cash, because the risk of an option is related to the risk of the underlying stock. Traders use the model not to predict the future, but to interpolate from a liquid stock and a liquid bond to an illiquid option price. It works pretty well but not perfectly, because its naļve assumptions aren’t quite true.
Now the question: What are the roles of models in identifying and managing risk?
Let’s look at the principles of taking risk. Loosely, there are two kinds of risk taking:
(i) making naked bets on directions, of stock prices or volatilities or default rates, and
(ii) making hedged bets, as is done, for example, by derivatives traders or market makers.
You Need Relative Value Models for Hedged Bets, and That’s OK for the Private Sector.
Relative value models are relatively good for identifying and managing the risk of hedged bets, because hedges are relative positions. These models are very useful for an individual desk in estimating exposures. I’m all in favor of it for use by a particular desk or a particular product area of a bank that is making markets or carried out hedged trades. These are reasonably good models for the private sector. Reasonable is good enough. You have to expect to lose money sometimes.
In the Public Sector You Need Absolute Value Risk Models for Naked Bets, And They Don’t Work Well At All.
But the really big risks that can take down a firm or a financial system come from naked bets or poorly hedged relative bets. The public sector requires protection from naked bets, and modeling their risks requires forecasting and absolute models, and those, as I said, suck. (In my experience most traders who make naked bets don’t use complicated economic models or even listen very much to the economists in their banks.) So I don’t think any of these models are really good at managing risk.
Furthermore, the shattering risks to firms and systems come from illiquidity and contagion. No one quite knows what liquidity really is, and so there’s no good model of it. (Liquidity is a metaphor based on fluids. People define it by its proxies – bid-ask, average daily volume, market impact etc, which are features of but aren’t actually liquidity itself.)
So What Does The Public Sector Need?
Academics and perhaps regulators sometimes think that just around the corner is the “right” model that is going to work correctly. That’s not the case. So what can one do?
When I attended risk committee meetings at Goldman ten years ago, I was impressed but how well people understood the limitations of models and what to do about it. They didn’t live by mechanical bounds on risk according to some model. They understood that danger comes in many forms, that there is no generic tail risk, and that you need to protect a firm as a whole by putting many heuristic bounds on your exposure: to creditors, to leverage, to countries, to liquidity, etc. They understood that stale positions are probably dangerous, and therefore should probably be unwound. I think that’s the right way to proceed, to use models but no be aware that they aren’t gospel, to keep thinking about what can happen inside of and outside of your model, because the world isn’t a model.
I think that’s the right way to go about risk from the public sector point of view. The public sector needs to attract skilled experienced capital markets people who aren’t beholden to where they were trained and don’t intend to go back there, because their incentives as public protectors will have to differ from the private sector.
A Principle: If You Use a Model, You Are Short Volatility
All models are analogies, and being analogies, they are limited in their scope. In physics you can describe ice, water and steam, and the phase transitions between them, with one unified theory, amazingly, and hence you can handle the extremes of freezing and boiling.
In finance or economics we have nothing like that. Even beautiful Black-Scholes-Merton ignores volatility variations, illiquidity, panic, government regulations on shorting, to name just a few things that lie outside it.
Therefore, when the world changes dramatically, every single model you can think of is likely to fail. I would like the following principle to be engraved on the foreheads of all financial and economic model users: All models are short volatility. When volatility changes a lot, the model is going to fail.
A second question I was asked to consider is: Does Public Disclosure of Models Make Them Less Useful? Does disclosure result in a competitive disadvantage, and if so, does the public benefit outweigh the costs?
I think many people on Wall Street justify many things on the grounds of secrecy and efficiency, and most of this is just a knee-jerk defence of making a profit at all costs. Models are important, but more important is computerized risk systems that can aggregate over the entire firm. Building these are labor-intensive projects. Therefore I think that public disclosure of risk models is valuable and important; it doesn’t reduce competitiveness that much; its public benefit to the taxpayers who are on the hook clearly outweigh any disadvantages. Similarly I believe regulators should have as much access to linkages between firms, to exposures, to protect both the taxpayers and the firms and even the banks who will suffer if the worst happens.
Capitalism: Some Principles
As I’ve argued, there is no model that will tell you when catastrophe is coming.
(I am nevertheless hopeful for some of the agent-based models that take account of collective effects and regime changes. They are only models but they may suggest metrics related to the number of links between participating agents in the economy that give warnings of risk and danger.)
Nevertheless, what we really need is a few strong defensible principles that, if rigorously applied, will produce incentives that mean less regulation is needed. Unfortunately, these principles have been violated very badly during and after the great financial crisis. Let me list a few that I think are good. I’m sure there are more, but still less than the number of regulations.
- If you want the benefits of risk taking, you must suffer the disadvantages too.
- Don’t treat (only some people’s) insolvency as illiquidity.
- Efficiency isn’t everything. Efficiency alone isn’t a good reason for anything.
- You can’t solve political or spiritual problems by tackling the money supply. You can only postpone them.
- If you stimulate at some times, then you must dampen at others.
- “If you believe that capitalism is a system in which money matters more than freedom, you are doomed when people who don’t believe in freedom attack using money.” (Edward Lucas)
- Above all, do no harm.