Are Government Spending Multipliers State Dependent? Evidence from U.K. Historical Data Submitted in partial fulfillment of the requirements for distinction in the degree of Bachelor of Arts (B.A.) in Economics Gabriel P. Fritsch Advisor: Aleh Tsyvinski Yale College New Haven, Connecticut April 3, 2017 Abstract A recurring question in macroeconomic policy debates is whether government multipliers are higher during periods of economic slack or when interest rates are at or near the zero lower bound. This paper aims to shed further light on this issue by expanding the recent literature pioneered by Ramey (2011), which uses the defense narrative approach to identify shocks in the expected present value of government spending. Using newly constructed quarterly historical data for the U.K. from 1910 to 2016, I estimate multipliers that are below unity irrespective of the amount of slack in the economy or the monetary policy regime. Overall, there is some evidence that multipliers are higher during periods of high unemployment in the U.K. but no evidence of larger multipliers at the zero lower bound. 1 Introduction With the onset of the Great Recession in 2009, questions regarding the effectiveness of spending stimuli and the merits of austerity returned to the forefront of the macroeconomic policy debate in developed economies faced with persistently high unemployment and the need for eventual fiscal consolidations. This discussion has most recently been revamped in the United Kingdom following its decision to leave the European Union and the subsequent deteriorating economic outlook due to the uncertainty surrounding the exit process, with the UK Treasury signaling a shift towards easier fiscal policy.1 Questions concerning the magnitude of the government spending multiplier and whether it changes according to different states of the economy are central to this debate. The majority of estimates in the empirical literature based on aggregate data find modest multipliers, mostly below unity, suggesting that increases in public spending are ineffective in stimulating private activity, or, alternatively, that austerity measures are unlikely to significantly harm growth. The majority of these estimates are taken from averages for a specific country over a certain historical period as aggregate government spending multipliers can only be obtained by exploiting “natural experiments” in national data, given the impossibility of conducting randomized control trials across different countries. While the theory suggests that several characteristics can significantly alter the magnitude by which fiscal changes affect growth – namely, whether these changes are permanent or temporary, the method through which they are financed, the level of resource utilization in the economy, or how monetary policy responds to such changes – aggregate economic data is subject to endogeneity problems that make answering these questions a challenging exercise. 1 See “With Philip Hammond’s Plan, U.K. Shelves Austerity” The New York Times, 23 November, 2016. Web. The empirical literature has recently begun to explore whether multiplier estimates vary according to the amount of slack in the economy (e.g. Barro and Redlick (2011), Auerbach and Gorodnichenko (2012, 2013), Fazzari, Morley, and Panovska (2015)) or to different monetary policy regimes (e.g. Miyamoto, Nhuyen and Sergeyev (2015)). The first question attempts to answer the traditional Keynesian assumption that multipliers are high during deep recessions, as output is demand-driven and government spending acts as a substitute for subdued private activity. The second question is a more recent research interest that has arisen as a response to theoretical papers which use New Keynesian DSGE models to suggest that government spending multipliers are higher when interest rates are at or near the zero lower bound (e.g. Christiano, Eichenbaum, and Rebelo (2011)). However, Ramey (2011) highlights important methodological issues in the burgeoning empirical literature on state-dependent multipliers, by showing that the established identification method for spending shocks through a standard vector autoregression (SVAR) misses their timing. The paper proposes a novel approach to measuring anticipation of spending shocks through narrative methods of defense spending, where news sources are used to compute changes in the expected present value of government spending previously unanticipated by the economic agents. Using this method, subsequent studies (Owyang, Ramey, Zubairy (2013), Crafts and Mills (2013), Ramey and Zubairy (2014)) estimate government spending multipliers across both slack states and monetary policy regimes. Owyang et al. (2013) and Ramey and Zubairy (2014) construct defense news series for the U.S. since 1889 and for Canada since 1912, estimating multipliers below unity for both slack and non-slack states, with no evidence of larger multipliers during higher unemployment periods for the U.S. and some evidence for Canada. Ramey and Zubairy (2014) finds mixed results for multipliers at the zero lower bound, with some specifications indicating multipliers as large as 1.5. Crafts and Mills (2013) also constructs a defense news series for the U.K. from 1922 to 1938, which includes prolonged periods where interest rates are near the zero lower bound, finding multiplier estimates below unity (between 0.5 and 0.8). However, the paper does not estimate separate multipliers for ZLB and non-ZLB periods. This paper aims to extend the small empirical literature on the state-dependence of government spending multipliers using defense news shocks by constructing new quarterly data on estimates of changes in the expected present value of government spending for the United Kingdom from 1910 to 2016, thus expanding the narrative series in Crafts and Mills (2013). As argued in Owyang et al. (2013), covering the entire 20th Century contains potentially richer information than the post-WWII sample commonly used in previous empirical research (or the interwar sample used in Crafts and Mills (2013)), as it includes episodes of large variations in government spending, wide fluctuations in unemployment, and different periods with interest rates near the zero lower bound. Using the new quarterly series, I investigate whether government spending multiplier differs across slack states and monetary policy regimes in the United Kingdom. As in Owyang et al. (2013) and Ramey and Zubairy (2014), multiplier estimates are obtained through state-dependent models using Jordà’s (2005) local projection method. I estimate multipliers that are below unity irrespective of the amount of slack in the economy or the monetary policy regime; most estimates are between 0.3 and 0.8. Overall, there is some evidence that multipliers are significantly higher during periods of high unemployment in the U.K. but no evidence of larger multipliers at or near the zero lower bound. The paper proceeds as follows: Section 2 describes the data and considers the motivations and challenges of constructing and using the historical sample for the United Kingdom; Section 3 describes the econometric methodology and key issues in estimating multipliers; Section 4 presents the baseline estimates for multipliers across states of economic slack and robustness checks to the baseline specification; Section 5 does the same for multipliers across monetary policy regimes; Section 6 concludes. 2 Data In this section, I discuss the advantages and challenges involved in using historical data for the United Kingdom to estimate government spending multipliers, which is followed by a brief description of the data construction process. 2.1 Advantages and Challenges of Using Historical Data Since it is impossible to conduct randomized control trials across countries to measure the effect of changes in government spending, researchers have to resort to identifying “natural experiments” in aggregate historical data, where large, exogenous changes in government spending allows for the extraction of its effect on the economy from the many other contemporaneous confounding shocks. As discussed in Ramey and Zubairy (2014), this becomes even more challenging when a state-specific investigation is involved, as it requires a significantly large sub-sample containing exogenous shocks to spending. To overcome this challenge, historical data for the U.K. is constructed dating back to 1910, which includes several periods of economic slack with deep recessions and high unemployment, and two prolonged periods of interest rates at or near the zero lower bound. Using data covering most of the 20th century also accounts for the two largest increases in government spending in the history of the United Kingdom as a result of both world wars. A valid question, however, is whether the estimates using historical samples offer informative conclusions for modern policy: if the U.K. economy has changed significantly over time, estimates for government spending multipliers using data for the entire 20th Century are perhaps not indicative of what multipliers have been in recent years. Ramey and Zubairy (2014) respond to this issue by arguing that if multipliers were to have changed over time, they would intuitively be smaller, not larger, as better financial market access and greater consumer sophistication have led to less hand-to-mouth consumers (i.e. consumers who do not save or borrow), and thus a lower overall marginal propensity to consume. Additionally, monetary and fiscal stances are fairly well