The challenge of stress-testing liquidity risk

Date: March 31, 2015

The liquidity risk paradox, inspired by the FT

David Oakley highlighted in the Financial Times (13th March) how top investment groups are shock-testing their bond portfolios for the increased liquidity risk that is being perceived by the main financial actors in the bond market.

Increased regulation is forcing banks to reduce the amount of capital dedicated to their market-making operations and offload proprietary trading activities that were often synergic to those operations. As a consequence, bond markets are becoming thinner and liquidity is shrinking, especially for non-core issuers.

Another increasing pattern is that market depth tends to disappear during moments of stress. When the wind changes all market operators seem to be on the same side of the market, offers proliferate and bids disappear. Other factors contribute to this erosion:

  • Absolute investment strategies have increased their market penetration in recent years, with the consequence that the number of market factors that promise to avoid losses in downturns are increasing: they tend to sell all at the same time unfortunately;
  • Risk regulation is making the market safer from many perspectives, but since almost all risk measures tend to be pro-cyclical, they exacerbate market corrections, triggering risk breaches during downfalls and adding the number of operators that sell, when markets head south.

For all the above reasons, it is not a surprise that top investment groups worry about liquidity risk. The irony is that this time the poor risk manager is confronted with a nasty problem. The less a bond is liquid, the less information he has for measuring liquidity risk. Very illiquid issues actually have no data at all, so the paradox is that where you mostly need data to measure your liquidity risk, you have none.

Read: New Insights for Asset Managers: How Technology Can Drive the Most Effective Middle Offices

I defined this as the liquidity risk paradox some time ago: no matter how smart and quantitatively elegant your model is, it will fail systematically and with direct proportionality to the illiquidity of your assets. The more they are illiquid, the less useful your model is.

Stress testing is therefore a good approach and a better alternative to traditional risk models when measuring liquidity risk, especially when bonds and derivatives are concerned. But how should we stress test liquidity? Not every bond will behave in the same way, bonds of a same issuer will be more liquid in shorter maturities, higher grade bonds maintain liquidity better than low-grade issuers and so on so forth.

So, the first condition for a comprehensive stress test is to identify all the main characteristics that differentiate the liquidity behavior of a bond in moments of stress. There are some obvious parameters that stand out:

  • Maturity: by definition a bond that matures in the next few days will be highly liquid. Maturity is a first obvious parameter to take into consideration; longer bonds will be, ceteris paribus, less liquid than shorter bonds;
  • Credit Grade: it is a fact, high-quality issuers are more liquid than low-quality and they maintain liquidity during stress exponentially more than low grade issuers;
  • Currency Denomination: the currency in which a bond is denominated is another key dimension of liquidity risk. Bonds defined in emerging market currencies are unavoidably less liquid than bonds defined in advanced economies’ currencies;
  • Number of Risk Drivers: most bonds are just priced out of the reference yield curve and the additional credit spread that they offer. However, more complex bonds are exposed to additional risk drivers, like swaption volatilities (i.e. callable, CMS bonds) or stock prices and their volatilities (i.e. convertibles). The more the risk drivers involved, the more the hedges that a market maker has to apply, the worse the liquidity of the bond;
  • Bond Seniority: this is another obvious factor of liquidity, but it is normally reflected inside the Credit Grade, already presented above.

This list is already long and may be non-exhaustive. Now, think of the relations among these factors: is a high-grade long-maturity bond in USD more or less liquid than a high-grade ZAR bond but short term? When you think of all these factors you quickly realize that the possible combinations of questions like this makes a stress test implementation quite difficult.

Is there a way to simplify this framework? Can we represent all the complexity of these relations inside a solid and intuitive quantitative framework?

Risk drivers are not only used to price bonds but also for immunizing their positions from market movements. Good hedging also reduces their capital charges. Risk drivers used for hedging are instruments like Interest Rate Swaps or Credit Default Swaps. Their prices are observable and they proved to be more resilient than bonds during the last crisis, when liquidity is concerned.

A solid liquidity stress test can be built by first designing scenarios of bid/ask spreads for the risk drivers. These scenarios can easily capture that, in situations of stress, longer maturities, lower credits, less liquid currencies widen the bid/ask spreads of the respective risk drivers proportionally to the level of stress.

At this point, we can capture in a quantitative way, how those risk drivers’ bid/ask interact inside each individual bond: the pricing function of the bond does that job for us. We first find out the sensitivity of the bond to each risk factor. For instance if a market maker is hit on his bid for a convertible bond, he will be long stock, long stock volatility, long credit and interest rate risk. He wants to go long on those risks profitably, which means that he will select the best bid/ask for each risk driver used for hedging. In our example, he will use the bid of the stock and stock volatility and the ask of IRS and CDS to quote his bond bid. A bid on a reverse convertible, instead, would imply using the ask of stock volatility, as the sensitivity is reversed in that case.

This approach has the benefit of covering any asset, independently from its current liquidity status, and availability of data and to take into consideration all the main dimensions of liquidity risk under any possible scenario, inside a strong quantitative framework. The last five years of experience so far tell us this is a consistent and scalable approach to measure liquidity risk responding to the increasing challenges posed by today’s bond markets.

Looking for improvement in your middle office?