Abstract: Does borrowing diversity (i.e., borrowing via a larger number of debt types) affect how firms respond to an exogenous credit supply shock? To answer this question I use the recent 2007-2009 credit crisis as a negative exogenous credit supply shock to U.S. non-financial companies. Applying a difference-in-differences methodology, I find that during the crisis companies that ex ante borrowed from many debt types had significantly higher capital expenditures than otherwise similar companies that borrowed from fewer debt types. The former group also had higher market valuations, a lower cost of debt, a lower reduction in debt issuance, higher leverage ratios, and a lower need to use internal cash during the crisis. This evidence is robust to applying an instrumental variable estimation, which takes into account the endogenous nature of the diversity measure. Finally, further tests suggest that borrowing diversity could represent a valid measure for financial constraints.
Presentations: Georgetown University, University of Miami, Erasmus University Rotterdam, BI Norwegian Business School, ESADE, Federal Reserve Bank of Richmond, Cornerstone Research, Federal Reserve Board of Governors, EFA Doctoral Tutorial, FMA USA, SGF Conference, Australasian Finance and Banking Conference), SFI Corporate Finance Workshop, Northwestern Causal Inference Workshop, EFMA, Austrian Working Group on Banking, University of Vienna, VGSF Annual Students Conference
with Franklin Allen, Marlene Haas, and Eric Nowak
Abstract: On October 26, 2008, Porsche announced its domination plan for Volkswagen. This announcement came as a surprise to investors shorting Volkswagen stock, and caused a short squeeze that briefly made Volkswagen the most valuable listed company in the world. We use the Porsche-VW short squeeze and the German financial market system as a unique experimental setting to argue that regulation is important for market quality and informational risk in modern, dynamic, yet opaque financial markets. We provide the first forensic academic study of this squeeze and show that it significantly impaired price discovery, increased informational risk, and impeded market efficiency. These limits to arbitrage in the form of short sale risks imply significant costs to the arbitrageurs involved.
Presentations: Vanderbilt, University of Lugano*, ISM Munich*, the University of Luxembourg*, the National University of Singapore*
Abstract: We discuss a novel role for covenants and accounting-performance measures in credit lines. During aggregate liquidity shortages, banks need to ration liquidity. Credit line covenants allow a bank to revoke the credit line if a firm's accounting-performance measure falls below some threshold. Revoking credit lines protects banks against severe aggregate liquidity shocks. Idiosyncratic and transitory shocks in the accounting-performance measure have two benefits. First, they introduce randomness in covenant violations that eliminates concerns of favouritism when banks ration liquidity. Second, when transitory shocks are correlated with the level of aggregate liquidity shock, the likelihood of covenant violations after severe aggregate shocks is higher than in normal times, improving the allocation of liquidity. Implicit liquidity insurance can complement covenants, inducing banks to revoke credit lines of covenant violators only after severe aggregate shocks, not in normal times. Consistent with this revocation pattern, we find a positive association between covenant violations and credit line revocations in the crisis of 2007-2008, controlling for firm fundamentals, but not outside the crisis.
Presentations: Columbia Business School*, Cass Business School*, NYU Stern*, Temple University*, University of Washington*, VGSF, The Accounting Research
Workshop in Zurich*, European Accounting Association*, The Accounting Group Meeting of the German Economic Association*, EWFS*, FDIC/JFSR conference*, University of Melbourne*
with Josef Zechner and Jeff Zwiebel
Abstract: DCF corporate valuation usually features a terminal value to capture cash flows beyond the typical forecasting horizons of three to seven years. Despite its dominating effect on overall firm value, the academic literature provides very little guidance on how one of its main inputs, the long-term growth rate, should be determined. This paper presents an exploratory analysis of how firms’ long-term growth is related to various firm and industry characteristics. We apply an extensive selection of predictors and document a negative relation between long-term growth rates and variables representing market expectations, systematic risk, and firm age and size. We also find a positive relation between variables representing managerial perception and firms’ competitive positioning and subsequent long-term growth rates. Share prices do not seem to capture the full information in long-term growth. We find that a trading strategy that goes long
the decile with the highest long-term growth expectations and short the bottom decile yields positive and statistically significant abnormal returns in the range from nine to twelve percent per year.
Presentations: Vanderbilt, VGSF, Spängler IQAM
Gold Forecastability: An Evaluation of Model Performance
(paper available upon request)
Abstract: In this study I compare the in-sample and out-of-sample performance of several econometric models with respect to gold price forecasting. The models that I apply are: 1.) the simple random walk with a drift, 2.) the Multiple Linear Regression model (MLR), 3.) the Autoregressive Moving Average model (ARIMA), 4.) the Vector Error Correction Model (VECM), and 5.) the single equation Error Correction Model (ECM). The empirical results show that the MLR model has the best fit to the data as measured by the adjusted coefficient of determination and SBC. However during the one month out-of-sample forecasting period the ARIMA (2,1,1) model has the best performance as measured by RMSE, MAE and MAPE. During the three, six and twelve months out-of-sample forecasting periods the MLR has the best performance. Nevertheless, during the 24 months out-of-sample forecasting period the ARIMA (2,1,1) model outperforms again all other models.