Lawrence Klein, the Nobel Prize-winning pioneer of econometrics, died at his home on Sunday. I had the privilege of working at Wharton Econometric Forecasting Associates (WEFA), a commercial entity that had evolved from the original non-profit, Wharton Econometrics that Larry Klein had founded in 1963 at the University of Pennsylvania. Larry was still actively involved as an advisor to WEFA during that period. He trained and advised many students and employees of that firm over the years. Today, the legacy of WEFA lives on at IHS Global Insight which acquired it. >
Others will give a thorough review of his professional accomplishments, but let me share some recollections of our work together. Larry, by his own admission, wasn’t a “wunderkind” in mathematics. However, his application of applied mathematics and statistics to better understand how the macro economy works and in developing macro econometric models that could be used for policy analysis and forecasting was a highly creative endeavor. Jan Tinbergen had made an attempt in the 1930s, but Larry developed the first large-scale macro econometric model of the U.S. economy. It was enhanced and expanded over the years and extended to include virtually every major country in the world. Ultimately, this work led him to being awarded the Nobel Prize in 1980. He was constantly attempting to improve the ability of macro models to trace actual changes in the broad economy, and pushing others to do so—myself included.
I became the Director, U.S. Long-term Macro Forecasting at WEFA in the summer 1992. I inherited what was called the “Mark 9” version of the U.S. Macro model to project 25 years into the future and conduct policy analysis for private clients and government agencies at the federal and state level. That was just before the 1992 presidential election. Larry had been an advisor to the Clinton Campaign and after he was elected, assisted them on developing an economic policy agenda. At the direction of Larry, our WEFA U.S. macro team performed a number of policy simulations to see what might improve the performance of the economy. Many late-night model runs were performed. The model had been recently moved from the mainframe to the PC. To perform a 25-year alternative simulation of the U.S. economy, after hours of setting it up, the model would require 30 minutes to solve because of all the iterations and how slow PCs were at that time. My computer today could solve it in five seconds.
Larry was famous within the profession for his call in 1945 that the end of World War II wouldn’t cause the U.S. economy to go back into a recession, as many others had predicted. I had the opportunity to ask him how he had been so confident. Larry told me that there was substantial pent-up demand of consumer durables, cars in particular, among returning soldiers and consumers that would be satisfied. He told me that a model builder had to have the courage of his convictions. He built the model and had to believe it was correct or abandon it. I carry that advice with me to this day. His insight helped inspire me to examine the stock adjustment process built into the current equation for consumer durables. The properties were largely non-existent. I re-estimated the WEFA equations for personal consumption expenditures of all types of durable goods and they became part of the “Mark 10” version of the WEFA Macro model.
I was promoted to Senior Vice President in 1995 and moved over to head up the U.S. Regional Service at WEFA. In early 1996, Larry and I discussed my idea of a better way to predict presidential election outcomes based on economic performance. Ray Fair and others had been building models to predict presidential outcomes for may years; however, they were predicted at the national level. Presidential elections were actually determined by the electoral college which was based upon state voting patterns. I told Larry that we could do a better job using WEFA’s projections for state economies and predicting the electoral college. Larry encouraged me to proceed to test my hypothesis.
I developed such a model across all 50 states and the District of Columbia using data from the modern presidential election period of 1960 through 1992—nine presidential elections. Basically, the prediction model was based upon growth in state real wages per capita over the four quarters prior to the election, the level of state unemployment in the quarter prior to the election, and a national misery index (year-over-year inflation plus the prime rate). It included voter registration preferences in each state and whether a sitting incumbent was seeking re-election. People were generally predisposed to keep an incumbent in office. Voters usually afforded an incumbent a 2.5 percentage point advantage, holding economic performance factors constant. This model explained past voting behavior very accurately.
We used the WEFA economic projections for the second and third quarters of 1996 and tested the model to see whether the incumbent, Bill Clinton, would be re-elected. After the Republicans gained control of the House in 1994, people predicted that Bill Clinton was in jeopardy of not being reelected. However, the model predicted that Clinton would father 371 electoral college votes. He ultimately received more than enough to win.
Larry presented this prediction at an American Economic Association Meeting during August, 1996. He told me that there were some skeptics in the audience, but that he had great confidence in the model builder, me, and in the projection. That was an endorsement of which I was extremely proud and remain so. Later int he campaign, when it was clear that Bob Dole had a limited pathway to winning the election, the model was used to determine which states the campaign should focus on. I updated this model just last year. It predicted every state but one correctly for President Obama.
I saw Larry in 2000, at the Milken Institute’s annual Global Conference. At the dinner the night before, he asked me if I had updated the Presidential election model. I told him it suggested that the incumbent party, the Democrats, and Al Gore in particular, would face a very close election based upon the economic model prediction., I couldn’t have known that the election outcome would be determined by the U.S. Supreme Court.
Just this morning, I discovered that while working at the National Bureau of Economic Research in 1947 to 1948, Larry worked on a production function estimation focused on the railroad sector. I never knew that Larry had performed this work. It is important because I started my professional career at Union Pacific and later became Chief Economist at CSX, both major U.S. railroads. Larry used to ask me about how railroads operated. I thought he was just showing an interest in me, butI suspect he really wanted to know because he never completed his work on the rail production function. Maybe he thought that he might return to it. Larry Klein will be sorely missed.