Liquidity Premium Analysis Highlights Value in Real Estate ABS


By Christopher Flanagan, director, Ryan Asato, vice president, and Ralph DiSerio, first vice president, Merrill Lynch Real Estate ABS Research

With many spread products recently trading at or near spreads last seen during the crisis of 1998, value appears to abound across all sectors. The question which naturally arises is which sector offers the best value.

We start by recognizing that the current value opportunity flows from the market demand for liquidity. The investor willing to provide liquidity - or stated differently, take liquidity risk - has the opportunity to receive handsome incremental compensation above and beyond that already provided for prepayment and/or credit risk.

There are essentially two key aspects to the opportunity: 1) earning the liquidity premium for the duration of the bond holding period; and 2) realizing spread tightening, related to any eventual erosion of the purchased liquidity premium, by the end of the bond holding period. While the first component is essentially a sure bet (assuming we have already been compensated for the prepayment and/or credit risk), the second component is not.

Liquidity demands could actually increase in the future and spreads could move even wider. Thus, the "opportunity" lies in anticipating that an extended holding period will be required to allow full realization of both components of the liquidity compensation.

The next step in assessing relative value is to determine the liquidity premiums available in various products, as well as the potential for change in that premium. Unfortunately, measurement of liquidity premiums has not attained the degree of scientific advancement seen when measuring other risk factors, such as prepayment option cost.

Moreover, to the best of our knowledge, data on the liquidity premiums embedded in various spread products are not widely or readily available. To fill this gap, we provide what we believe is a simple yet viable approach to computing these premiums. The conclusion from the analysis is that the potential rewards from providing liquidity to Real Estate ABS make the sector particularly attractive at this juncture.

Using a Spread-Volatility Regression Model

To Determine Liquidity Premiums

The basic premise or assumption of our approach is that all spread product have an embedded option - a prepayment and/or a default option. Since an increase (or decrease) in volatility should increase (or decrease) the value of these embedded options, spreads should widen (or narrow) as volatility increases (or decreases). Assuming this, we perform a simple linear regression of spreads against some measure of volatility. Testing a number of different volatility measures, the best R-squareds for Real Estate ABS resulted from using the implied yield volatility of 3-month options on the 5-year Treasury.

We chose the observation period, 1/1/97 to the present, for two reasons. We felt that HEL ABS spreads had not fully "normalized" until the beginning of 1997, when market concerns about the convexity characteristics of HEL ABS were for the most part alleviated.

Furthermore, it was about this time that reduced Treasury issuance caused on-the-run Treasurys to trade increasingly rich to off-the-runs and we saw increased liquidity premiums reflected in the pricing of spread products.

When the spread volatility regression was performed, we obtained a reasonable linear fit of data. Nonetheless, standard regression analysis should include analysis of the error terms, and in a sense, this becomes the more interesting part of the analysis of the 5-year HEL regression.

Recognizing that liquidity premiums have been on the rise during the period, we tested the hypothesis that the model needed to incorporate a measure of liquidity premium in order to take out the autocorrelation.

By regressing spreads against both volatility and the 5-year Treasury liquidity premium (the difference between the Merrill Lynch spline model and the on-the-run 5-year Treasury yields), we found that the autocorrelation was eliminated and the R-squareds increased to above 90%. This suggested to us that the "problem" with the simple one-factor model is that it is missing a term which captures a liquidity premium.

To develop an estimate of the unique liquidity premium embedded in each of the spreads we considered, and to facilitate relative value assessment, we assumed that the error terms were, in fact, the liquidity premiums for each product.

If we were to keep the 5-year Treasury liquidity premium in our model, we would lose our ability to establish unique liquidity premiums for each product. This is a very important assumption. So we repeat: for each product, the liquidity premium is defined as the error term from the spread-volatility regression model.

Results