Working Papers
- Monetary policy surprises: robust dynamic direct and total causal effects (with Lynda Khalaf. Job Market Paper)
- Abstract: In this paper, we study the dynamic causal effects of a monetary policy shock on the US economy within the Local Projection - Instrumental Variable [LP-IV] framework. Our reassessment is motivated by the emerging concerns in the literature about popular IVs that are based on high-frequency identification. We approach related difficulties as follows. First we provide weak-instruments robust inference on the traditional LP-IV coefficient which we denote as the direct causal effect [DCE]. Second, we define, estimate and test an alternative response parameter, denoted as the total causal effect [TCE], that accounts for the inherent unobservable endogeneity factor resulting from the first stage regression error. The TCE is identified whether the considered IVs are weak or strong. Our view is that both effects play an important role in capturing the net impact of a policy shock. In the context of two baseline empirical models with factor controls, results confirm that conventional 2SLS methods produce statistically insignificant responses at conventional levels. Using identification-robust approaches produces economically more plausible results, yet overall, we find that instruments are weakly informative on DCEs. Estimates of the TCEs provide critical insights which, in sharp contrast to DCE estimates, are mostly unchanged as additional credit spreads are considered. We find that outcomes can go in the opposite direction from what theory would predict on credit markets and the macroeconomy, which suggests that DCEs may miss important responses.
- Revisiting the Flow View of Quantitative Easing: Evidence from Asset Purchases (with Hashmat Khan)
- Abstract: In this paper, we examine the flow view of quantitative easing (QE) using monthly data on Federal Reserve’s pre-announced asset purchases from the second and third rounds of QE. We determine both average and cumulative purchasing effects using structural VAR and local projection methods, respectively. For financial assets, we find that the purchasing shock increases the stock price index and the 10-year treasury yield, but it decreases the housing price index. For macro aggregates, there is no statistically significant average effects of the purchasing shock, however, we find that the accumulation of the purchases does increase both the industrial production and the consumer price index.
- Basel Liquidity Regulation and Bank Lending in the U.S. (with Lynda Khalaf)
- Abstract: In this paper, we examine the impact of the Liquidity Coverage Ratio (LCR) on bank lending in the U.S, using a Difference-in-Difference framework with a variety of identification methods. We are particularly interested in treatment effect dynamics. In this context, the dynamic two-way fixed effect (TWFE) model is commonly used which consists in including dynamic indicators for time relative to treatment which allows for treatment adoption to vary across time. The coefficients on these indicators aim to track the evolution of treatment effects. However, recent econometric works suggest that TWFE estimators do not recover the hypothesized causal effect; severe bias cannot be ruled out even when treatment effect dynamics are homogeneous and in the absence of anticipatory behaviour. The underlying reasons for such failures can be summarized as follows: (i) parallel trend assumptions (PTAs) - some of which may often be implausible - with varying strengths lead to different interpretable causal effects; (ii) the interpretable causal effect is in fact an (unknown) linear combination (weighted average) of the indicator coefficients, (iii) the role of conditioning covariates and treatment anticipation. Available evidence on the LCR ratio is scarce and is restricted to standard event studies. In this paper, we compare standard dynamic TWFE estimates to recently proposed alternative specifications that allow us to introduce various group-time aggregation schemes. Results underscore the importance of defining clear interpretable parameters, allowing for conditioning on covariates. In general, we find no effects of the LCR on bank lending, and the assumptions embedded in the TWFE models translate into meaningful differences in empirical results.