A Model of the Demand for British Government Stocks by Non-Bank Residents 1967-77

Abstract

This paper develops an econometric model of the demand for gilt-edged securities and other long term public sector debt by the non-bank private sector. We chose to work initially at this high level of sectoral aggregation since it is the one most commonly used in studies of the demand for money and the term structure of interest rates. The high level of aggregation permits comparisons with earlier work and also allows alternative theoretical hypothesies to be tested in a straightforward way. We started out by testing the simplest of these alternatives - the so called 'Expectations Hypothesis'. This suggests that the difference between the observed yields on short and long term bonds equals the capital gain expected on the latter. (See for example Roll (I970), p. 36 n.) The immediate empirical implication of this theory is that differences in these returns which are observed ex post, taking into account capital gains, are unexpected and therefore should not be systematically related to variables which are observable ex ante. However our work showed that such ex post relationships could be found suggesting that a more general hypothesis was necessary to describe the data. An alternative approach is to attempt to estimate the demand function in some appropriate way. The usual method, following the original work of Modigliani and Sutch (I966) is to invert the demand function and explain the yield gap in terms of expected capital gains and relative asset supplies. However such models of the term structure are only valid econometrically if the supply variables are in fact exogenous, an assumption which is unlikely to have been true in the United Kingdom. Moreover, Rowan and O'Brien (I970) have shown that a proper application of this approach should involve the imposition of a set of structural restrictions. These have an important influence on the results but are usually disregarded at the empirical stage. In order to avoid these problems we chose to estimate the demand function directly, using instrumental variables to handle interest rates and other current endogenous variables. Previous models of the term structure have assumed an autoregressive expectations formation process and allowed for this by introducing lagged interest rate terms directly into the regression. However, the resulting estimates * This model was originally developed for use in the financial sector of the Treasury macro model (Spencer and Mowl, 1978). I am grateful to Peter Middleton, Colin Mowl, Steven Bell and other colleagues for encouragement and helpful advice. The paper has benefited from comments by David Hendry, John Flemming, Marcus Miller and other members of the Treasury's academic panel as well as the referees of this JOURNAL. The views expressed in this paper are those of the author and not necessarily those of the Treasury.