Hedging Against Fed Surprises

Summary

• A new NBER working paper examines how Fed language during FOMC meetings affects the prices of key asset classes.
• The results imply the prices of the US 3m, 10y and 2s10s are more likely to respond to surprise hikes on FOMC announcement days than the VIX, credit spreads or DXY.
• Meanwhile, to hedge against Fed risk around Fed meetings, the authors recommend buying staples and growth sector industries. They typically offer lower returns on FOMC days, but higher returns on the days surrounding the meeting.

Introduction

Systematic analyses of central bank communications emerged in the early 2000s. The goal was to understand how language, both its tone and context, affected market prices. In the US, research has shown Fed communication influences markets in two ways. It does so in the short term by conveying information about monetary policy. And it does so in the long run by conveying news about the state of the economy.  

A new NBER working paper, co-authored by researchers from Chicago and Columbia, uses natural language processing (NLP) to predict how FOMC statements and minutes should affect asset prices. The difference between this Fed-implied value and actual values is a measure of what the authors call ‘the Fed communication surprise’, or FDIF. This surprise contains interesting properties: the higher it is for the VIX, credit spreads and the DXY, the worse the macroeconomic outlook becomes.

The Theory

In theory, market prices contain information from Fed communications because central bank signals lead investors to make inferences about the prevailing monetary policy stance and the state of the economy.

Take the VIX, for example. Its level is a function of S&P500 option prices. These prices move systematically during FOMC meetings, meaning they are impacted by Fed communication. By deduction, then, so is the level of the VIX.

Theoretically, if you can determine how Fed communication influences the VIX, you could predict where the VIX should go based on the language used in the latest FOMC meetings. This is what the authors try to do – not just with the VIX, but with the US 3m, US 10y, 2s10s, DXY, and HY and IG credit spreads.

Methodology

The authors map words in Fed statements and minutes to the value of each of the assets.

To get the words, the authors use Python (BeautifulSoup and nltk packages) to scrape statements and minutes from the Fed’s website and construct a list of words and bigrams for each document. After the texts are processed, they use word and bigram counts as a proxy for the Fed’s communication on each FOMC announcement day.

Next, they need to map how the tone of the Fed’s words relate to the market price of each asset. For this, they use a support vector regression (SVR). Very simply, it gives a quantifiable relationship between a word, e.g., ‘inflation’, and the asset price, e.g., US10y. In this example, a positive relationship means that the more inflation is mentioned, the higher the US10y is predicted to go.

Once a set of relationships between words (roughly 7,500) and an asset price is uncovered, the authors can create a Fed-implied market value of that asset price on the day of each FOMC announcement. It turns out that this value is highly correlated with actual market prices, with correlations ranging from 65-95% (Chart 1).

Digging Deeper Into Fed-Implied Values

The Fed-implied values and actual market prices will never be 100% correlated. One reason is because actual market prices on FOMC announcement days will reflect more than just information contained within the Fed statements and minutes (‘noise‘).

Another is the Fed mentioning new words during major events (‘model surprise’). In Chart 1, there are occasions where the Fed-implied value lags the actual market prices, like in 2008 and 2020. This is because the Fed starts using words that the model has not been exposed to yet, such as ‘outbreak’, ‘coronavirus’, ‘crisis’, and ‘quantitative easing’.

It takes at least two successive meetings for the model to learn how to attribute these words to the price action of different asset classes. The reason it takes time for new relationships to emerge is because the analysis is run over rolling windows of up to 20 FOMC meetings.

SVR analysis produced a list of important words for the three asset classes in Chart 1 (Table 1). Words like inflation and labour market positively impact rates, while growth and employment tend to be associated with a lower 10y and 2y.

Decomposing the Fed Communication Surprise, FDIF

We now have the Fed-implied market price and the actual market price. The difference between the two is called ‘the Fed communication surprise’, or FDIF. Why would there be a difference between the two?

Well, there is the ‘noise’ and ‘model surprise’. But the former is less significant on FOMC days. And the latter is small outside major economic events.

That leaves only a third component, which the authors call the ‘macro information surprise’. Here, a surprise comes when the Fed releases information on the state of the economy, or announces a policy choice, that differs from market expectations.

State of the economy surprises, the authors find, are more likely to show up in the prices of non-rate-based measures, like the VIX, credit spreads and DXY.

Meanwhile, monetary policy surprises are more likely to reflect price movements in the three rate-based measures (3m, 10y, 2s10s) on FOMC meeting days.

FDIF in Practice

Below, we have the FDIF measure for three of the seven asset classes (Chart 2). The 10y and 3m, which move procyclically, are right-skewed, i.e., they have more upward jumps, while the 2s10s is left-skewed. More upward jumps simply means the Fed-implied price is typically higher than the market price.

FDIF and the Macroeconomy

The authors find forecasters predictably change their expectations based on Fed language. This suggests the FDIF contains important macroeconomic information.

For example, a larger difference between the Fed-implied VIX and actual VIX (i.e., a higher FDIF) from one quarter to the next is associated with a drop in corporate profitability forecasts.

A higher FDIF is also associated with negative revisions of economic growth forecasts and positive revisions of inflation. The same is true for credit spreads, and the reverse is true for US 3m and 10y.

What does this mean? Well, the VIX and credit spreads are countercyclical – they are high in bad times and low in good times. Their FDIF measures are associated with negative growth revisions. These, in turn, are associated with dovish revisions in expected monetary policy and therefore upward revisions in inflation expectations.

So, a higher FDIF for countercyclical assets is bad news for the macroeconomy but good news for procyclical assets, such as rates.

Fed Risk Premium

Lastly, the authors analyse the extent to which Fed risk is priced in markets. In other words, they investigate the risk premia associated with the cross-section of FDIF exposures during Fed information events.

If the FDIF measures capture important macroeconomic news, either about the state of the macroeconomy or about monetary policy, that are not known to the market beforehand, exposure to Fed risk should be priced.

It is. The authors find that countercyclical FDIF measures (VIX, credit spreads, DXY, US 2s10s), which are associated with negative macroeconomic revisions, command a negative risk premium. In other words, investors are willing to accept lower returns from holding hedging securities heading into Fed announcement days. The opposite is true for US 3m and 10y.

Why? FOMC announcement days are associated with important macroeconomic news releases. Securities which allow investors to hedge against bad news act as insurance and offer low average returns.

The hedging securities that do well on Fed announcement days when there are positive FDIF values on the VIX, credit spreads, DXY or US 2s10s are called ‘Fed-safe industries’. These industries have positive (negative) loadings on FDIF variables with a negative (positive) price of risk. And they typically perform worse than Fed-risky industries on Fed announcement days but better on all other trading days.

Fed-safe industries are staples and growth sectors, while Fed-risky ones, which have the opposite behaviour, are highly cyclical and are exposed to commodity or housing-related businesses.

Being able to identify when FDIF is high or low for particular assets on FOMC announcement days could be profitable. If an investor knew the FDIF on the VIX was high, and they went long Fed-risky industries and short Fed-safe industries on announcement dates, they could earn an annual risk premium of 7.5%.

Bottom Line

Albeit a slightly abstract paper, it reaffirms an important point: FOMC meetings are key events for investors, especially statement releases (the minutes less so). A measure that extracts the tone of these minutes and overlays them with historical price actions does well at predicting forecasters’ behaviours: if the Fed’s text-implied price is significantly different from the market’s price for non-rate-assets, expect worse macroeconomic outcomes over the next quarter. To hedge against this on FOMC announcement days, consider finding industries that typically perform well FDIF is high. The authors recommend staples and growth sector industries.