Abstract:
This talk illustrates two ways of combining theory and reduced form statistical techniques to make causal inference. The first example considers the effects of taxes on corporate leverage. Using variation in state corporate income tax rates, we find, contrary to prior research, that corporate leverage rises after tax cuts, mostly for small private firms but to a lesser extent for public firms. We use an estimated dynamic equilibrium model to show that tax cuts result in more distant default thresholds and thus lower credit spreads. These effects outweigh the loss of the interest tax deduction and lead to higher optimal leverage choices, especially for firms with stringent default conditions, such as the small private firms we study. The second example considers the use of a model to understand causal inference about a variable that does not yet exist: central bank digital currency. We estimate a dynamic banking model to quantify the impact of a CBDC on banks. Our counterfactuals show that a one-dollar introduction of CBDC replaces bank deposits by 80 cents on the margin. Lending falls by 25% of the drop in deposits because banks partially replace lost deposits with wholesale funding. This substitution raises banks' interest-rate risk exposure, lowering their resilience to negative equity shocks. If CBDC bears interest or is intermediated through banks, it captures a greater deposit market share, amplifying the impact on lending. CBDC especially affects small banks, which face expensive wholesale funding.
Bio:
Toni Whited is the Frederick G L Huetwell Professor and Professor of Economics at the University of Michigan. Professor Whited received her B.A. in economics and French, summa cum laude, from the University of Oregon in 1984 and her Ph.D. in economics from Princeton in 1990, working with Ben Bernanke. Professor Whited has taught in a wide variety of areas in finance, macroeconomics, and econometrics at the undergraduate, MBA, and doctoral levels. She has published over 40 articles in top-tier economics and finance journals. Her research covers topics such as the effects of financial frictions on corporate investment, econometric solutions for measurement error, corporate cash policy, structural estimation of dynamic models, monetary policy, and corporate diversification. She has won a Jensen Prize for one of the top articles in corporate finance in the Journal of Financial Economics and twice won a Brattle Prize for one of the top articles in the Journal of Finance in corporate finance. She is the past-president of the Western Finance Association, and she serves as editor-in-chief for the Journal of Financial Economics.
Summary
Randomization is the gold standard for inferring causality in complex systems
But… we almost never get randomization in reality
And when we do, the randomization is rarely informative about the underlying mechanisms
Economists use theory to justify causality
Causal Graphs (Pearl style)
Verbal justifications or event studies designs
Need a model of the underlying system to enable causal inference to be actionable and inform action in the real world
Taxes and Leverage
Question: how do corporate taxes affect firm leverage
High debt means low taxes
But also increases risk of default
So need to maximize debt upto bankruptcy
Data: US Federal Reserve tracks debt data on firms (private/non-public):Y14 collection
Borrower financials from 37 largest banks since 2011
Model:
Event study: tax changes
Linear model with 3 leads and 4 lag time points
Predicts change in debt
Results: corporate debt increases after tax cuts and decreases after tax hikes
Strong for small/private
Small for public
None for large private
Dynamic equilibrium model
Firms are financed by internal profits and external risky debt
Hire, invest and adjust debt
Interest expense is tax deductible
Model dynamics:
Tax shields make debt more attractive
But,
Taxes make firms less profitable/valuable
Lower the threshold at which the firm defaults
This causes lenders to charge higher interest for loans to these now more vulnerable borrowers
Estimated model parameters using the simulated method of moments (any sampling-based optimizer would work)
Output parameters: estimated elasticity of leverage to staggered tax changes
Simulated: .041
Real: .021
Statistically not differentiable
Used a best-fit model to make predictions about firm behavior across different tax rates, firm debt loads, etc.
CBDC
Central Bank Digital Currency
Makes digital coins more reliable
Makes it easier for people to pay, since banks are not involved
Makes the banking system less involved (less control, less credit supply, less money in the banking system)
No public data on deployments of CBDC, so used a model based on the real economy and did causal inference within the model
Used dynamic industry-equilibrium model to predict outcome of CBDC on deposits
Banks maximize distribution to shareholders
Take deposits/make loans
Financial frictions and imperfect competition
US has few large banks
Banks give better loan terms to larger banks
Imperfect substitution between banking products for customers (cash and bank deposits are very different)
Data: Federal Reserve Call Report data
Estimated sensitivity of demand to different attributes of banking product attributes (convenience, rates, security)
Used separate causal linear models
Introduced CBDC into model
New bundle of financial product characteristics (more convenient, etc.), based on existing CBDC deployments
Outcome:
CBDC can replace a significant fraction of private money creation (~80%)
More if they pay interest
Banks become less profitable but then replace deposits with loans from money market funds at a less profitable rate