The Study in a Nutshell

There has always been a question of how hedge funds and asset managers make money. By exploiting a comprehensive regulatory database, a new Bank of England working paper finds that these investors have made abnormal returns in bond markets since 2011. They have done so by obtaining a competitive edge from observing other investors, responding quickly to macroeconomic news and correctly forecasting macroeconomic fundamentals.

The chart below captures the predictive power of fund managers concisely. It shows the event-time cumulative returns of the long-short portfolios (long on the top tercile and short on the bottom tercile of government bonds) sorted by daily order flows of the two investor types. We can see that hedge fund trading positively forecasts bond returns in the short run followed by a strong reversal in the subsequent month. Mutual fund order flows, on the other hand, positively forecast bond returns in the subsequent two months.

Charts 1 & 2: Fund Managers Are Good at Forecasting Bond Returns

Source: Page 44 of ‘Informed Trading in Government Bond Markets’

The paper also shows that hedge funds and mutual funds have a significant advantage over other market participants in collecting, processing, and trading on information that is relevant for future gilt returns. In particular, the findings highlight the differences in the two groups’ approaches to earning abnormal returns in the government bond market.

A. HFs gain from both trading ahead of other investors and quick responses to the arrival of macroeconomic news.

B. MFs profit from their ability to understand and forecast macroeconomic fundamentals.

Through their active trading, these professional managers help to impound value-relevant information into gilt yields and expedite the price discovery process in one of the world’s most important financial markets.

The Theory of Bond Yield Variations

There are two competing theories the literature puts forwards for variations in bond yields:

I. The traditional view: Monetary policy announcements and the arrival of public information drive the main source of variation in the terms structure of interest rates. According to this view, trading in government bond markets is mostly due to rebalancing and hedging needs and is unlikely to have a large, persistent effect on bond yields.

II. The alternative view: Heterogeneity in investor beliefs generate variations. This stems from differences in investors’ access to information and their ability to relate publicly available economic fundamentals to the term structure of government bond yields. An immediate prediction of this view is that as long as learning is imperfect, trading of the better informed should persistently outperform that of the less informed.

The BoE working paper focuses on the second channel. A large empirical literature on institutional trading has so far found little evidence that professional money managers are able to earn significant abnormal returns in stock and corporate bond markets. Instead, this research asks whether a subset of non-dealer institutions have superior knowledge about future government bond returns.

On this front, the authors find:

‘Daily hedge fund trading positively forecasts gilt returns in the following one to five days, which is then fully reversed in the following month.’

‘Mutual fund trading also positively predicts gilt returns, but over a longer horizon of one to two months. This return pattern does not revert in the following year.’

The work then goes on to address why fund trading forecasts government bond returns, and for this they build on recent theoretical work by Farboodi and Veldkamp (2019). They postulate that arbitrageurs can engage in two types of activities: (i) to predict and trade ahead of other investors’ demand, and (ii) to learn about future asset values in an accurate and efficient manner (more so than the average investor in the market).

Both mechanisms are examined, and the authors find:

‘Part of short-term return predictability is due to hedge funds’ ability to anticipate future demand of other investors’.

‘It is partly due to mutual funds’ ability to forecast changes in short-term interest rates’.

The Data

The study uses the ZEN database, a comprehensive regulatory dataset maintained by FCA containing all secondary market trades in UK government bonds (gilts) by all FCA-regulated financial institutions. Given that all gilt dealers are UK-domiciled and hence FCA-regulated institutions, the ZEN database effectively covers the entire trading activity in the UK government bond market.

The database offers three main advantages:

1. It provides detailed information on all individual transactions (the date and time stamp, transaction price, transaction amount, etc.).
2. One can observe the identities of both counterparties in each transaction (for example, a transaction between a dealer bank and a bond fund).
3. It covers nearly all investors and transactions; more precisely, the buy and sell transactions in our sample sum up to the total trading volume in the gilt market.

The sample period spans August 2011 to December 2017. They only keep bonds with a time-to-maturity longer than one year and exclude inflation-indexed gilts from the sample. The granularity and completeness of the data enable them to systematically analyse the extent to which any investors have a competitive advantage in this market and, furthermore, are able to profit from their information edge.

The final sample consists of 55 gilts covering nearly all gilt transactions. The majority of guilt trades take place in the inter-dealer market. The following chart shows the market share in the UK government bond market.

Charts 3 & 4: Hedge Funds and Mutual Funds are Big Players in Bond Markets

Source: Page 43 of ‘Informed Trading in Government Bond Markets’

For the investigation of whether funds are able to use value-relevant public information efficiently, the authors focus on announcements of UK inflation and labour statistics, and the Monetary Policy Committee (MPC) meetings. MPC meeting dates are collected from the Bank of England, and the UK Office for National Statistics publishes the other macro-announcement dates.

Finally, to calculate risk-adjusted bond returns, they construct three tradable factors mimicking the level, slope, and curvature factors of the term structure of government bond yields. For the level factor, they use the value-weighted average return of all available gilts. For the slope factor, they use the return differential between the twenty-year gilt and the one-year gilt. The curvature factor is the average return of the twenty-year and one-year gilts, minus that of the ten-year gilt.

The Results in Detail

This section is divisible into three parts. One, results on the predictability of HF/MF trading on gilt returns. Two, the ability of HFs/MFs to predict other investors’ returns. Three, the ability of HFs/MFs to learn from value-relevant information and respond to it more efficiently than other market participants.

1. On predictability of bond yield returns, the authors sort gilts (with different maturities and vintages) into terciles based on the previous-day net purchases of HFs/MFs. This is because the market does not immediately and fully respond to the order flows of HFs and MFs, so one would expect to see a price drift in the same direction in subsequent periods.

They find that both HFs and MFs have significant information advantages in the gilt market. There is a strong positive correlation between HF/MF trading and contemporaneous gild returns – gilts heavily collectively bought by hedge funds and mutual funds on a particular day outperform those heavily sold by 1.82 bps. Specifically:

Hedge Funds

• The tercile of gilts heavily bought outperform the tercile heavily sold by 1.28 bps on the following day, and 2.88 bps in the following week, with an annualized Sharpe Ratio of 1.2.
• The return spread then becomes a statistically insignificant 1.32 bps (t-statistic = 0.73) by the end of month one, and -1.28 bps (t-statistic = -0.31) by the end of month two.
• This return predictive pattern is virtually unchanged after controlling for known risk factors (i.e. the level, slope, and curvature factors). For example, the five-day three-factor alpha of the long-short bond portfolio remains economically and statistically significant at 2.94 bps.
• Consistent with these results based on daily order flows, monthly hedge fund order flows have no predictive power for bond returns in the subsequent months.
• Finally, these results also hold in Fama-MacBeth regressions (used to control for omitted variables, such as lagged bond returns and known predictors of government bond returns) and exhibit strong persistence in the cross-section of hedge funds.

Mutual Funds

• The return spread between the top and bottom terciles of gilts (long-short portfolio), sorted by the previous-day mutual fund order flow, is a statistically insignificant 0.45 bps on the following day, and an insignificant 1.75 bps in the following week.
• The return spread then grows to 6.47 bps by the end of month one, and to 15.61 bps by the end of month two. There is no evidence of reversal over the following twelve months; the cumulative return of the long-short gilt portfolio by the end of month twelve is nearly 1.3%.
• Sorting gilts into quintiles based on the previous-month mutual fund order flow, they find the return spread between the two extreme quintiles in the following month is 27.52 bps, with an annualized Sharpe Ratio of 1.5.
• After controlling for known risk factors, the three-factor alpha is only modestly reduced to 17.98 bps per month. This return pattern again exhibits strong persistence in the cross-section of mutual funds.

2. The authors regress order flows of hedge funds in the same bond in the previous week onto aggregate order flows of an investor type (mutual funds, non-dealer banks, and ICPFs) in a bond in the next five days. They find:

Hedge Funds

• Daily trading is a strong predictor of future mutual fund trading; a one-standard-deviation increase in hedge funds’ net buying in a week forecasts an increase in mutual fund net purchases in the following week by more than 1%.
• Hedge fund trading is largely unrelated to future order flows of non-dealer banks and ICPFs; and, importantly, aggregate order flows of other investor types (aside from hedge funds) do not predict future order flows of hedge funds.
• Isolating the part of mutual fund trading that can be relatively easily predicted (capital-flow-induced trading), they find that hedge fund order flows significantly and positively predict mutual funds’ flow-induced trading in the following week.
• This means hedge fund trading should be more profitable in periods of relatively large mutual fund flow-induced trading in absolute terms – they are. The long-short gilt portfolio sorted by hedge funds’ order flows earns significant abnormal returns only in periods with high aggregate absolute FIT.

Mutual Funds

• MF trading (measured at the daily or monthly frequency) has no predictive power for future order flows of other investors, consistent with the view that mutual funds are usually not specialised in forecasting the demand of other investors.
• Instead, the authors link the trading activity of mutual funds to future movements in the term structure to identify whether mutual funds are able to forecast variations in certain parts of the yield curve – they can. Shifts in the weighted-average portfolio duration significantly and negatively forecast changes in short-term interest rates (the one-year rate) one to three months in the future.
• For example, at the three-month horizon, the coefficient on changes in mutual funds’ average duration is a statistically significant -1.73. This estimate implies that a one-standard-deviation reduction in the average portfolio duration of mutual funds forecasts a 4.49 bps increase in the one-year interest rate.

3. Finally, the paper repeats the return predictability test of HF/MF trading separately for macro-announcement days and non-announcement days. Again, they sort all gilts into terciles based on hedge fund order flows on the day prior to the announcement. They then track the performance of the long-short portfolio on the announcement day. In a time-series regression setting, controlling for known predictors of future interest rates, they find for:

Hedge Funds

• The long-short portfolio sorted by hedge fund daily trading earns substantially higher returns on macro-announcement days. Hedge funds earn nearly twice as much on announcement days (2.50 bps) than on non-announcement days (1.28 bps).
• Interestingly, hedge funds seem to earn higher abnormal returns on labour/inflation statistics announcement days than on monetary policy announcement days: the long-short gilt portfolio sorted by hedge fund trading earns an abnormal return of 1.22 bps on MPC announcement days vs 3.53 bps on inflation/labour statistics announcement days.
• On this, the authors write: ‘A potential explanation for this result is that labour/inflation announcements contain less forward-looking information than monetary policy announcements, consistent with the short-lived outperformance of hedge funds’.

Mutual Funds

• Out of the 17.98 bps monthly alpha earned by mutual funds, 7.24 bps are earned on just two days: one with monetary policy announcements and the other with inflation and labour statistics announcements.
• Put differently, mutual funds earn 3.62 bps/day on macro-announcement days and only 0.5 bps/day on other days. These results suggest that about 40% of the total monthly alpha (7.24 bps out of 17.98 bps) are realized on just two macro-announcement days.

Bottom Line

Both hedge funds and mutual funds are informed investors in the gilt market. The former have short-term predictive power which can be attributed to their trading ahead of other investors’ predictable order flow. Mutual funds also positively predict bond returns, but over a horizon longer than one to two months. The superior performance of mutual funds is partly due to their ability to forecast future movements in short-term interest rates.

The punchline: nimble hedge funds are good at trading ahead of other investors’ future demand; mutual funds are instead more concerned with economic fundamentals.

To view the full paper – click here