How banks can avoid bad haircuts on hedge fund trades

Banks that finance hedge fund trades know a thing or two about bad haircuts. A new paper from a quant at HSBC suggests they try a different style.

When financing hedge funds, banks have long stuck to the financial equivalent of the old short back and sides. Stocks purchased with borrowed cash are assigned a lower-than-market value and posted as collateral in the transaction. The reduction in collateral value – the so-called haircut – is typically based on the value-at-risk of the underlying assets. As such, it largely depends on the volatility of the assets and time horizon of the trade, though the final figure is often subject to intense negotiation.

The problem with this approach is that it neglects the creditworthiness of the hedge fund. This may seem redundant when dealing with inherently risky counterparties, but including credit metrics in the calculation could prove useful in certain circumstances – for example, when trading repos and total return swaps that allow recourse to the borrower.

The stock return is modelled with a double exponential jump diffusion model, a flexible approach that captures up and down jumps of different magnitudes and frequencies

Wujiang Lou, director of quantitative trading at HSBC in New York, proposes a new model for calculating haircuts and funding rates that incorporates an equity and a credit component.

The stock return is modelled with a double exponential jump diffusion model, a flexible approach that captures up and down jumps of different magnitudes and frequencies. The model accounts for asymmetries consistent with those observed in markets, and is especially adept at capturing short-term equity moves over the margin period of risk, typically three-to-five days.

Credit risk can be modelled as a standard log normal distribution or using the Black-Karasinski model, which assumes the log of the default intensity is mean reverting.

To connect the credit risk and market jump diffusion, Lou needed to calculate the correlation between them. This would be relatively easy for counterparties with traded credit default swaps. That’s not the case with hedge funds. To overcome this hurdle, Lou used indexes.

“I started looking at the correlation between CDX and the S&P 500, and it is stable,” he explains. “We use a factor model in which we assume a hedge fund’s CDS is a component of the CDX index. On the stock return side, each stock will be modelled with a jump diffusion model which is related to the S&P 500 index.”

The index correlation is used as a proxy for the correlation between a hedge fund’s default risk and its equity holdings.

Lou argues the collateral haircut and the funding rate paid by hedge funds should be determined jointly, and introduces a repo pricing model to calculate the latter.

In practice, banks would set the desired credit rating for a financing deal. Using that as an input, Lou’s model outputs a target level for the haircut. Once the haircut is agreed, the funding rate is calculated to ensure the bank’s exposure is in line with the desired credit rating.

For the bank, a bigger haircut reduces contingent exposure, allowing for a lower funding spread. The problem for hedge funds is that this will increase their overall cost of funding, as they will need more of it. The trick is to use the economic capital that hinges on the haircut to optimise funding costs. The bigger the haircut, the lower economic capital the bank needs to set aside. Lou suggests collecting a charge on this economic capital usage if a hedge fund wants a lower haircut to reduce its overall funding costs.

Running the model on a sample portfolio of the stocks traded by Archegos Capital Management would have resulted in significantly bigger haircuts than were applied by Credit Suisse and some of its other counterparties – to the point where the positions would have been prohibitively expensive to put on at that size.

While Lou’s model is viable for calculating haircuts on hedge fund trades, he wants to see more modelling and research into the topic – especially by credit quants. “Because repos are considered simple products, generally credit quants are not involved in modelling their risk profiles and pricing,” he says. “A common misconception that many people in the industry have is that repos have no recourse to the borrower, while that is actually the case. Therefore, there is indeed additional credit support in the event of default, which in principle should allow to reduce the haircut applied to the borrower.”