The recent rise in inflation has put the Federal Reserve back in the spotlight. How should monetary policy be conducted in the current economic environment? To answer this question, one must understand the causal effects of monetary policy on the economy. Since the Fed adjusts its policy based on a change in economic outlook, this is not straightforward. One approach that macroeconomists have adopted is to look to past data and to isolate interest rate changes that are not a reaction to economic conditions, but rather arguably external.
Romer and Romer (2004) suggest measuring outward movements at the rate of federal funds because of the difference between observed and intended changes in rates. Objective changes are based on the economic outlook of policy makers when making their decisions. Romer and Romer (2004) use numerical forecasts of inflation, output and unemployment in the ‘Greenbook’ document prepared by Fed economists for Federal Open Market Committee (FOMC) meetings. This method is applied in subsequent studies, for example Tenreyro and Thwaites (2013) and Coibion et al. (2014).
In a recent study (Aruoba and Drechsel 2022), we propose an innovative approach to detecting the impact of monetary policy. We follow the concept of data exploitation in documents prepared by Fed economists for the FOMC. Our approach, however, aims to capture more extensively the information contained in these documents, including numerical predictions as well as human language. We do this with natural language processing and machine learning methods.
A new way to detect the impact of monetary policy through language
From the forecast of changes in the federal funds rate, we infer the monetary policy push between (i) and (ii). (ii) To obtain, we first identify the most important economic terms in the document. This results in a set of 296 single or multi-word expressions, such as ‘inflation’, ‘economic activity’, or ‘labor force participation’. Figure 1 shows a word cloud of 75 frequently mentioned economic concepts between 1982 and 2017. The size of each concept reflects the frequency across the document
Figure 1 Frequently Asked Economic Concepts, 1982-2017
We then create sentiment indicators that capture the degree to which these economic concepts relate to positive or negative human language. The basic idea is to calculate the number of positive or negative expressions based on a pre-defined dictionary, which appears close to each economic concept. Our collection of 296 sentiment time series paints a rich picture of the historical assessment of economic conditions by Fed economists. Figure 2 shows the perception of the vicinity of ‘economic activity’ as an illustration. This time series reflects the variables of the meaningful business cycle, shrinking sharply in the recession.
Figure 2 Feelings surrounding the concept of ‘economic activity’
With regression in the federal funds rate a regression to the left and (i), (ii), and (iii) to the right is unlikely because there are far more regressors than observations. To overcome this problem, we use machine learning strategies: we employ a ridge regression to predict targeted changes in federal funding rates using our huge set of regressors. The concept of a ridge regression is the sum of the remaining squares and an additional term that punctures the square deviation of each regression coefficient from zero.
Systematic vs. external changes in interest rates
Most economists would argue that monetary policy is governed by a very systematic approach. An interesting feature of our approach is that, in light of this realization, only a small fraction of interest rate changes are attributed to external shocks. Our ridge regression implies that the systematic component of monetary policy explains the 76% variability of the target interest rate, where 24% of the variables are responsible for the push. Compared to the existing application of Romer and the concept of Romer, the methodological element is significantly more important in our method. A broader set of forecasts, the sentiments of Fed economists, as well as nonlinearity all contribute to capturing more systematically the monetary component of monetary policy.
In our paper, we also examine whether the inclusion of additional information in our ridge regression changes our push measurements, using data from Fed Transcript and FOMC staff formation. We see that our push measurement has not been explained outside of the information provided to FOMC members by Fed staff at the beginning of a meeting.
Marked shock inspection
The dark blue line in Figure 3 plots the series during our approximate monetary policy push. The figure compares it to a light orange line with Romer and the approximate remnants of Romer. Our monetary policy shocks measure a generally low volatility and a low degree of automatic correlation. These are not just a scale-down version of the shocks implied by the original Romer-Romer method. In many cases, the orange line perfectly refers to a large thrust, while at other times the large thrusts are visible for the blue line.
Figure 3 Monetary policy push time series
For those episodes where the push for monetary policy is particularly large, we take a closer look at the discussions held at the FOMC. It focuses on capturing approximate monetary policy shocks. The largest external facilitation is estimated for the November 7, 1984 meeting of the FOMC. It is a period with a mixed economic outlook: industrial production fell for the first time in two years, yet investment and spending rose sharply. Fed staff concluded that “the slowdown could only be a break in a recovery that did not run its full course.” When we read the transcript of the FOMC meeting, it became very clear that many of the participants felt very optimistic about the forecast. Their policy actions are consistent with a major simplification of policy, providing a great example of a situation where the FOMC’s views on economics differ from those of staff economists. It is important to emphasize that this is an unusual situation. If disagreements occurred more frequently, our method would have highlighted them and predicted a change in policy.
The biggest external tightening occurred at the November 15, 1994 meeting. Fed workers argue that the consequences of inflation above the economy’s full potential have not yet been realized. They offer two policy options: a no-change option and one where the federal funds rate increases by 50 basis points. During the FOMC meeting, Chairman Greenspan suggested that since the market has already built in a significant rate of growth “a slight surprise would be of significant value.” He proposed an increase of 75 basis points to get “before the general expectation”. While most participants agreed with the proposal, several participants emphasized the credibility of controlling inflation. Again this is a situation where the FOMC has decided to take a step not only on the basis of the current economic outlook but also on other considerations. Our approach therefore implies that it reflects the shock of a monetary policy.
The impact of monetary policy hurts the economy
With our innovative measurements of monetary policy push in hand, we study the Impulse Response Functions (IRFs) of macroeconomic variables in a sophisticated Bayesian vector autorigration approximately from October 1982 to 2016. The results of a financial austerity are presented. The two panels in Figure 4 show the bonds yield, stock price, actual GDP, GDP deflator and IRFs of additional bond premiums based on our shocks (blue) and shocks created using the original Romer-Romer method. (Orange).
Figure 4 The impact of monetary policy hurts the economy
Using the push of our monetary policy, we see that a monetary austerity is consistent with the forecast of economic theory, leading to a decline in production activity and a fall in price levels. This is in stark contrast to the IRF with the shocks created from the original Romer-Romer specification, where a financial tightening does not seem to have a significant impact on economic activity. Previous investigations have already suggested that more recent samples push the IRF into monetary policy in disagreement with the theory, as discussed in Ramey (2016). One explanation is that some systematic policy variations may still be present in purely constructed shock measurements based on numerical predictions. Figure 4 indicates that the fancy method we develop can overcome this problem by incorporating a larger set of data.
Finally, our shock measurements do not appear to be subject to the ‘Fed Information Effect’ (Nakamura and Stenson 2018). Jarocinski and Karadi (2018) argue that a monetary tightening would raise interest rates and lower share prices, while misleading positive central bank data shocks both increase. Figure 4 indicates that our shock measurements lead to an increase in interest rates and a fall in stock prices. We conclude that natural language processing and machine learning are effective in providing a clearly identifiable estimate of the shocks of monetary policy.
Aruoba, B and T Drechsel (2022), “The Shock of Monetary Policy: A Natural Language System”, CEPR Discussion Paper No. 17133.
Coibion, O, Y Gorodnichenko, L Kueng and J Silvia (2014), “Innocent Pedestrians? Monetary policy and inequality ”, VoxEU.org, 25 October.
Jarocinski, M and P Karadi (2018), “Policy and Economic News Transition in US Federal Reserve Declaration”, VoxEU.org, 03 October.
Nakamura, E. and J. Steinson (2018), “High-Frequency Detection of Financial Non-Neutrality: Data Impact”, Quarterly Journal of Economics 133: 1283–1330.
Ramey, VA (2016), “Macro Economic Shocks and Their Promotion”, Handbook of macroeconomics 2: 71-162.
Romer, CD and DH Romer (2004), “A New Measure of Monetary Shock: Derivation and Implications”, American Economic Review 94: 1055-1084.
Tenreyro, S and G Thwaites (2013), “Pushing on a String: US Monetary Policy Less Stronger in Recession”, VoxEU.org, 12 November.