Federal regulators have begun exploring a new method of financial modeling, called agent-based modeling, which inches them closer to an optimal understanding of the true risk within the banking sector.
The only thing is that agent-based modeling still doesn’t address the core problem within a financial crisis: disorderly liquidation of investment positions.
In a must-read column in last Friday’s New York Times, regulators describe how agent-based modeling “tries to analyze what each agent — in this case, each bank or hedge fund — will do as a situation develops and worsens.”
Stress tests take a negative set of economic assumptions and ask how each bank would fare in those circumstances. As such, their usefulness is constrained by how well the assumptions reflect something that might actually happen. If it has never happened before, there is at least some chance that a stress test would not even consider what could be a severe problem.
This is absolutely true, and is the major shortcoming of any financial model. Financial models work only when everything — and I mean everything — is behaving within a relatively normal band of performance. In other words, a bank modeling its credit card portfolio creates a scenario whereby chargeoffs climb 10% or 20%, or maybe even 30%. But does the model accurately assess what happens if chargeoffs climb by say, 2,000%? No.
Here’s the technical explanation of the shortcoming, according to the Office of Financial Research’s working paper on agent-based modeling:
A key drawback of both VaR (value-at-risk) and typical stress testing practices is that they are guided by history. The volatility and correlations for VaR are drawn from an historical sample, and stress test scenarios typically replicate historical events or express extreme “tail events” based on an historical distribution even though it is well known that the nature of crises is to have unanticipated shocks and unexpected interrelationships where the past offers limited guidance. … The effect of a shock on a vulnerability in the financial system—such as excessive leverage, funding fragility, or limited liquidity—creates a radical shift in the markets similar to what is observed in traffic jams or the panic of crowds. Economic relationships change during these times of stress. Thus, the extreme event reflects the inappropriateness of the risk model, not an extreme draw from it.
The Office of Financial Research argues that agent-based modeling can work, because the “interactions—the attempts by agents to express actions within their world—are determined dynamically by the agents’ rules, which encompass the agents’ preferences and data-processing methods, and the information the agent receives about the state of nature. After each time step, the new environment and values for the variables resulting from the interaction of the agents is fed into the simulated world for the next-step iteration. The model can be run repeatedly, in effect creating multiple worlds, in order to generate a distribution of results.”
In other words, if you know how, say, a hedge fund will react to a financial crisis, you can model out that reaction and see how it dominoes through the banking world.
There are limitations to this. For example, one hedge fund might be leveraged more than the next, and that underlying, heightened leverage will lead Hedge Fund A to behave differently than Hedge Fund B. The authors of the Office of Financial Research paper understand this, and they say that agent-based modeling will allow regulators to better calibrate responses, such as raising the borrowing costs of Hedge Fund A to prevent its leverage from getting too out of whack.
In technical terms:
The insight that might be gained through this is the so-called volatility paradox, which is the tendency for volatility to drop when market risk and leverage are rising, luring investors into complacency. A bottom-up approach to this procyclical dynamic would model the effect of individual agents’ actions on volatility and thereby model volatility as an endogenous part of the market. This might uncover clues to the potential for a crisis. For example, it may help answer whether the low volatility resulted from increasingly levered agents who are supplying liquidity based on small price advantages.
So here is my problem with this: it is not the financial entity that is the problem, but the market itself. In other words, do we care whether Hedge Fund A or Hedge Fund B blows up or do we care about the sort of financial contagion we saw during the credit crisis? I would argue (actually, I have already argued this in 2009 here) that financial contagion is driven by extreme concentrations of financial holdings. “Extreme concentrations” do not have to be represented by large amounts of money. As anyone who has ever owned a penny stock knows, the dollar amounts might be small, but if one party holds too much of the overall investment pool, pricing can go haywire — and fast.
This is why regulators should focus on better understanding when an investment position will, if exited rapidly, cause a disorderly liquidation. Here is how Boaz Salik, my co-author, and I put it back in 2009:
Research we have conducted shows that an entity that holds more than five times the amount of shares or assets traded daily may lead to a disorderly liquidation should it need to offload its holdings quickly. We inferred this by looking at groups of publicly traded equities, which showed price volatility often increased substantially when the ratio of total shares outstanding to average daily trading volume grew too large. One likely reason is that the liquidations of those large positions often ended up collapsing the price of the public shares by creating a supply-demand imbalance.
The folks at the Office of Financial Research are starting to get this.
The model should allow for the range of shocks that are typical in causing and propagating a crisis. These include a seizing up of liquidity; a fire sale in the face of forced deleveraging with the subsequent funding and liquidity effects; a sudden funding impairment, which is often brought on by a shock to real or perceived creditworthiness or liquidity; or in the extreme case, the failure of a firm posed as an exogenous event.
But I would suggest that they spend more time thinking about the dynamics of deleveraging within a specific market, say the market for credit-default swaps, rather than what will happen at a particular financial enterprise during a market shock. By doing so, regulators will be able to avoid even getting to the point where particular financial enterprises are putting the financial system at risk.