Tinni Sen

Predicting a Country’s Growth: A First Look

In this chapter, we run different algorithm techniques to identify the algorithms that best predict growth. We show how machine learning can be used to validate different growth models. We suggest that validated algorithms enhance the confidence academics should place on any given theoretical growth model. We then show how machine learning can help researchers understand what kinds of concepts may make theoretical growth models more complete.

Predicting Recessions: What We Learn from Widening the Goalposts

In this chapter, we move our focus from economic growth to trying to predict a related target—economic recessions. We continue to use the “usual suspect” growth variables to check whether these variables are better at predicting recessions. We show how prediction performance of algorithms differs widely depending on the type of prediction criteria. We can, however, identify some of the most salient predictors of recessions. These suggest that fiscal policy may generally be better at combating recessions. Moreover, these predictors have non-linear effects on the likelihood of recessions which suggests that there may be no silver bullet for combating recessions. Last, the sorts of variables that influence economic growth also influence the likelihood of a recession. This suggest that economic growth probably should not be studied separately from recessions.

Why This Book

In this chapter, we lay out our plan for the book and the argument for the advantages of machine learning. ML algorithms use out of sample validation to identify which variables matter most for economic growth and recessions without making any heroic assumptions about the underlying distribution of the variables. Thus, it avoids problems of endogeneity. It also helps with the missing data problem by imputing missing data with validated algorithms. However, most of all, these algorithms can eliminate variables that are unlikely to be causal. Policy makers can therefore get a sense of the most important policy levers to change the path of growth.

Predicting Economic Growth: Which Variables Matter

In this chapter, we delve deeper into our findings in Chap. 4. We highlight how machine learning algorithms can highlight variables that have little predictive value relative to others. This machine learning technology can therefore help highlight the most salient growth “theory” among many. We also notice that the most predictively salient variables affect economic growth in a way that suggest equilibrium shifts in strategic models rather than smooth neoclassical patterns. Thus, we argue that machine learning approaches can help researchers identify more appropriate theoretical modeling techniques. Last, we suggest that some variables are better policy levers than others.

Data, Variables, and Their Sources

In this chapter we describe the data and how we prepare the data for analysis. We choose “usual suspect” variables that are widely known as the correlates of growth. Specifically, we look at Xavier Salaa I Martin’s work in identifying these variables. In the process, we include variables suggested by multiple theoretical growth models. We also show how non-parametric approaches can help create variables that capture the latent underlying factors that span multiple variables. This process also helps reduce unsystematic measurement errors.

When Good Little Debts Went Bad: Civil Litigation on the Virginia Frontier, 1745–1755

The use of a multinomial logit model to analyze a hitherto unavailable dataset of 1,376 small-claims lawsuits in colonial Augusta County, Virginia, for information about debts, litigants, and procedures f inds no evidence of prejudice in the legal system. The magistrates’ consistently fair enforcement of legitimate contracts may have induced both plaintiffs and defendants to settle their disputes in court rather than in private. The evidence corroborates the view that by the mid-eighteenth century, Virginia’s frontier judicial system was sufficiently impartial to encourage creditors to draw up efficient contracts even for small debts.

Price Dynamics and Asymmetric Business Cycles under Mixed State and Time Dependent Pricing Rules

This paper considers an optimal pricing model in continuous time that combines state and time dependent elements usually examined separately in the literature. In this model we find that recessions and booms are of roughly equal amplitude, contrary to results in Ball and Mankiw (1994) and Conlon and Liu (1997). On the other hand, while the amplitudes of booms and recessions are similar, their lengths differ. Applying the intuition developed in Ball and Mankiw to our model indicates that firms raise prices less frequently during recessions but more frequently during booms, so price-setters respond to booms more quickly.

The National Purchasing Manager's Index as a Predictor of Ex Ante Real Interest Rates - A Short Note

We look at the role of National Purchasing Managers Index in predicting real economic activity as captured by the real interest rate. We find that information about this index has the potential to improve forecasts of real interest rates, and therefore improve estimates of the state of the real economy.