Julia Giese, Michael Joyce, Jack Meaning and Jack Worlidge
Every financial market transaction has two parties, each with their own preferences. One channel through which quantitative easing works rests on these differences: preferred habitat investors value certain assets above others for non-pecuniary reasons, beyond risk and return. Central bank asset purchases of the preferred asset create scarcity, which may lead to compensating price adjustment, with spillovers to other assets and the macroeconomy. There is, however, little hard evidence on these investors. In a staff working paper, we use a new granular data set on gilt market holdings and transactions to identify groups of investors with preferred portfolio duration habitats. We present a case study suggesting that the Bank’s purchases appear to have come disproportionately from one group of these investors with a relatively strong preference for specific gilt maturities.
A unique data set
Until now the only way preferred habitat investors have been identified is indirectly, by assumption or by inference based on the behaviour of market prices.
Our approach is different. We take advantage of a unique data set provided by Euroclear that allows us to identify preferred habitat investors directly from their behaviour. This data set gives a near-comprehensive view of holdings and trades in the gilt market for our sample. It contains end-of-day gilt portfolios and high-frequency gilt market trades of accounts in the CREST system for each specific gilt. It covers a two-year period between 4 January 2016 and 31 December 2017, during which there were 9.8 million observations across days, accounts and specific gilts, and 3.4 million trades. By combining the stock and transaction information for those accounts associated with individual investors with publicly available information on the specific gilts held, we are able to construct a range of different measures for each investor portfolio through time.
We associate preferred habitat behaviour with minimising fluctuations in the average portfolio duration of their gilt holdings. We use a clustering algorithm to identify statistically differentiable investor groups based on the degree to which they maintain a stable weighted average duration of their gilt portfolio through time. The procedure models the data based on the assumption that observations are generated from one of J underlying multivariate normal distributions. This procedure allows for the possibility of multiple groups, but does not require there to be multiple distributions in the data. The resulting clusters classify investors into distinct groups, some of which more closely display the behavioural properties that theory associates with preferred habitat investors (see Chart 1).
In our benchmark analysis, four groups of investors, which account for a relatively large proportion of bond holdings in our sample, exhibit varying degrees of preferred habitat behaviour focused on different segments of the yield curve: one at the shorter durations (ST PHI in the chart), two at medium durations (MID PHI and MID2 PHI) and one at the longer end (LT PHI). The three other investor groups identified exhibit much larger variation in their portfolio durations, meaning they care less about holding the duration of their portfolio constant and consistent with ‘arbitrageur’ behaviour (investors who are purely motivated by risk and return considerations; ST ARB, MID ARB, LT ARB).
Chart 1: Clustering of investors based on the 10-90 range of portfolio duration and mean portfolio duration
Notes: Results from GMM algorithm estimated over 2016-17. Point size is scaled by average quantity of investor gilt holdings.
Who are the preferred habitat investors?
For nearly half the sample, it was possible to match an individual account with the underlying investor by using another data set. Further analysis on this part of the data suggests that the preferred habitat groups we identify include the investor types often associated with preferred habitat behaviour: foreign central banks, pension funds and insurers. What our data allow us to see is that not all preferred habitat investors are the same though. Foreign central banks are present at the shorter end of the yield curve; pension funds on the other hand tend to target duration habitats of 15 years or greater; with insurance companies sitting somewhere between the two (Chart 2).
Chart 2: Sectoral mapping of investor groups
Notes: Point size is scaled by average quantity of investor gilt holdings.
Through further testing of the behaviour of our different investor groups, we uncover a number of other features of identified preferred habitat groups, which both support our interpretation of these investors as akin to the preferred habitat investors of theory, and also illuminate their behaviour in practice. More specifically: they hold proportionately more of the stock of gilts; trade less frequently; and turn over their balance sheets more slowly than other investors.
An important theoretical feature of preferred habitat investors is also that they are less sensitive to relative price movements. In order to uncover this feature in our data, we regress the net change in an investor’s holdings of a particular bond on a fitting error for the specific bond, ie the deviation of the observed yield from a value implied by a statistical model. This fitting error is interacted with a set of dummies indicating whether or not a particular investor belongs to each of our seven previously identified groupings. Our results show that, as a bond becomes cheaper or dearer relative to the curve, investors respond by changing their holdings of it by more. However, investors that are in groups that our cluster analysis identifies as having tight preferred habitats are significantly less sensitive to the relative cost of the bond than investors in groups identified as arbitrageurs, that is preferred habitat investors are less sensitive to relative price movements than other investors.
A case study
Following the UK referendum on leaving the EU in June 2016, the Bank of England announced a package of monetary policy actions on 4 August 2016 to stimulate the economy, including a fourth round of government bond purchases (QE4). Between August 2016 and March 2017 the Bank of England purchased £60 billion of conventional gilts as part of this new round, taking the total stock of QE purchases to £425 billion. These gilt purchases provide an interesting case study for understanding the investment behaviour of preferred habitat investors in response to a shock to net bond supply. In an accounting sense, the Bank’s purchases would have been matched by sales from other agents in the economy, or an increase in the total stock of gilts outstanding. If the Bank’s purchases came from relatively price insensitive preferred habitat investors, they may have viewed the bank deposits they received in exchange as an imperfect substitute and looked to rebalance their portfolios into assets closer to those bonds. This ‘portfolio rebalancing’ would have led to an increase in the demand for other assets and thus a more generalised increase in asset prices and reduction in yields.
We can examine this episode using our estimates of the gilt holdings of different investor groups to produce a simple accounting of the counterparts to the Bank’s purchases between August 2016 and March 2017. Comparing the observed changes in gilts holdings to what might have been expected had the reaction been proportionate to the relative stock holdings of each investor group suggests that the Bank’s purchases seem to have come to a much larger extent than expected from the MID2PHI category of preferred habitat investors. As far as we can identify, these are more likely to be insurance companies with a portfolio averaging around 10 years in duration. The decline in holdings of preferred habitat investors seems consistent at face value with a wider portfolio balance channel (such as found in earlier QE episodes, see eg Joyce et al (2017)), although information on where these investors invested instead and a plausible counterfactual would be necessary for a full assessment.
By confirming the existence of preferred habitat behaviour for gilts, we provide empirical support for theories of QE that stress the potential importance of local supply effects: where central bank asset purchases reduce market yields by creating scarcity in sectors where there is strong but somewhat inelastic underlying investor demand. Our finding that preferred habitat behaviour exists across the term structure, rather than being restricted exclusively to longer maturities, may also have broader implications for understanding price dynamics in the gilt market: it suggests that the impact of demand shocks from these investor groups may be more pervasive than previously thought and that local supply effects may exist across the curve. We see rich avenues for further research to understand this more fully.
Julia Giese works in the Bank’s International Surveillance Division, Michael Joyce works in the Bank’s Monetary and Financial Conditions Division, Jack Meaning works in the Bank’s Chief Economist ED Office and Jack Worlidge works in the Bank’s Markets Intelligence and Analysis Division.
If you want to get in touch, please email us at firstname.lastname@example.org or leave a comment below.
Comments will only appear once approved by a moderator, and are only published where a full name is supplied. Bank Underground is a blog for Bank of England staff to share views that challenge – or support – prevailing policy orthodoxies. The views expressed here are those of the authors, and are not necessarily those of the Bank of England, or its policy committees.