Lecture Notes The narrative approach to establishing causation in
macroeconomics (Christina D. Romer – Keynes Lecture)
Dfn: empirical technique where sb gathers systematic evidence from contemporary
qualitative sources (newspapers, government reports, policy meeting transcripts) and incorporates it into statistical analysis Issues: 1. What are the strengths and weaknesses of the narrative approach 2. What is required to make narrative work scientifically rigorous and replicable? 3. How does it compare with other approaches? Does monetary policy matter? Yes it does! Consider regressing output (or unemployment) on changes in the policy interest rate It matters why the central bank is changing interest rates If countercyclical or anticipatory movements in interest rates are common, the coefficient on interest rates will be biased towards 0 (a large fraction of monetary policy is geared towards offsetting factors caused by other powers Policy makers may misjudge the importance of their actions Omitted variable bias is a ubiquitous problem in macroeconomics. Two other examples: The real effects of financial crises Causal relationship between the impact of government austerity programs and the economy Two steps in the narrative approach 1. Gather systematic evidence from qualitative sources (narrative analysis) 2. Incorporate it into a statistical framework for hypothesis testing Requirements for Strong Narrative Analysis 1. A consistent, contemporary, accurate narrative source 2. A clear idea of what one is looking for in the documents 3. Approach the source honestly and consistently Our narrative source Detailed minutes and verbatim transcripts of Federal Reserve monetary policy meetings For most of our sample, these records were expected to remain confidential Criteria for a monetary policy shock Contractionary: Policymakers chose to raise interest rates because they wanted to lower inflation, and were willing to accept adverse consequences for output and unemployment Expansionary: Policymakers chose to lower interest rates because they wanted to lower unemployment, and were willing to accept adverse consequences for inflation Contractionary Shock in December 1988 “I think the job before us is to contain inflation and to slow this economy down. Now, the danger is that I don’t think we have the time to do that…” Expansionary a Shock in January 1972 “The committee should consider whether stimulating the economy to greater heights in the short run would not involve a cost in the form of a resurgence of inflationary pressures later on” From 1940 to 2000, there have been 8 contractionary measures taken, recorded by the Federal Reserve and 4 expansionary measures Define a monetary shock dummy variable 1 in a month when there is a contractionary shock -1 in a month when there is an expansionary shock 0 otherwise We can scale our dummy variable to measure the extent of the contractionary or expansionary policy Contractionary shocks are more contractionary than expansionary shocks are expansionary (effects of expansionary shocks wane out more quickly) Monetary policy was a very important factor of recessions (USA) in the first 50 decades of post-ww2 period Contribution of monetary shocks to recessions Shocks account for more than 31% of USA’s recessions Other approaches to estimating the effects of monetary policy 1. Vector auto-regressions (VARs) 2. Residuals from a reaction function estimated on central bank forecasts 3. High-frequency identification Some other topics where the narrative approach could be useful 1. Institutions and economic growth 2. Financial structure or regulation and financial stability Is there an easier way to use the narrative approach? (More scientific and rigorous but less time consuming?) When are you able to delegate the narrative work to an untrained human or a computer? Machine Learning and Narrative Research Training is essential Even then, cases with limited numbers of observations or complex categorization may not feasibly be delegated to a computer