Working Papers
Working Papers
I study contagion effects from the European sovereign debt crisis into US wholesale funding market rates through the lens of granular portfolio data for US money market funds. I find that spreads between affected assets and other benchmark rates are rising by 30-40% during the European crisis compared with the remaining sample period (2010-2015). Using a large language model, I construct a text-based sentiment index to trace back contagion effects. The sentiment index explains more than 50% of the observed rise in US wholesale funding rates and its effect is more than twice as strong during the crisis period. These results are driven by money funds experiencing large redemptions that are only partially in line with actual portfolio risks, indicating that fund investors had limited or noisy information in making their redemption decisions. Finally, I also show contagion effects into long-dated US sovereign and agency debt held as repo collateral by money funds.
We use a detailed general equilibrium model of the euro area financial system to analyse the effects of fragmentation events in euro area sovereign bond markets. Using the May 2018 stress episode as an empirical case study, we show that the model matches empirical price and quantity movements closely. We show pro-cyclical portfolio re-balancing of non-bank financial intermediaries across different asset markets, while the banking sector acts as stabiliser in sovereign debt markets. In the model, targeted central bank interventions can be more than twice as effective in stabilising sovereign yields than broad-based quantitative easing programmes of equal size.
We construct an informationally robust monetary policy instrument (MPI) and use bank-level data on excess reserves for US commercial banks to analyze whether the emergence of significant excess liquidity in the banking system influences the transmission of monetary policy. For our empirical analysis we employ panel local projections instrumental variables (LP-IV) methods, instrumenting the effective federal funds rate with our MPI to determine the causal effect of the unexpected component of monetary policy. We further refine our analysis through a state-dependency, categorizing banks in our sample into “cash liquid” and “cash illiquid” banks, using bank-specific excess reserves data. Our analysis focuses on banks’ loan supply and differentiates between unexpected monetary expansions and contractions. Our results confirm the theoretical bank-lending channel for the pre-Great Financial Crisis (GFC) sample period for both types of banks, but show a significant heterogeneity once we include the GFC and post-GFC period. Namely, “cash-liquid” banks now react with loan-supply expansions to both unexpected monetary expansions and contractions, while “cash-illiquid” banks still react in line with existing economic theory.
(Work in Progress) (joint with Christian Weistroffer (ECB))
I combine granular data sets to develop an early warning system (EWS) for Euro Area (EA) investment fund redemptions during stress episodes. The EWS is based on using a hybrid machine learning model combining temporal features of a Long Short-Term Memory (LSTM) neural network with a Light Gradient-Boosting Machine to maximise the information extraction from a rich granular security-level panel data set of euro area investment fund portfolios. The model allows us to both study a rich set of potentially driving factors of fund flows and the estimation of sector-wide effects through the aggregation of individual flows.
Apart from being the first of its kind for non-bank financial institutions (NBFIs), a major contribution of this work is that we use a highly sophisticated graphical scraping process for proprietary Refinitiv Eikon data. This allows us to create a survivorship-bias free data set of EA investment fund portfolios covering the period from January 2010 to December 2023. We use the ECB’s Lipper Global Data Feed (LGDF) to employ complex string matching methods and retrieve ISINs for securities missing unique identifiers. We also identify financial derivatives in the scraped data set so that we identify in total more than 96% of all securities held by EA bond funds, equity funds and mixed funds. We further enrich the scraped and LGDF data with proprietary and ECB information at the security, security issuer, and investment fund level ( using the Refinitiv Eikon API, Refinitiv Lipper and the ECB's CSDB and SHSS). This information enables us to construct measures of portfolio risk, portfolio overlaps, the footprints of individual funds in different securities and markets, as well as general information on investment funds (including information on the holders of these funds).
The comprehensive coverage of EA investment funds, combined with the granular approach allows us to go beyond traditional econometric approaches and employ a significantly more holistic method capable of dealing with the high dimensionality of the problem we look at.
I develop a theoretical network model in which banks receive short-term liquidity from open-ended money market mutual funds (MMFs). These MMFs are used as an investment tool for cash pools, called investors. Banks are similarly modelled to Montagna and Kok (2016, ECB WP 1944) and as such have to satisfy liquidity and capital constraints and can further invest in risky, but illiquid, securities. They further hold interbank loans. Finally, banks hold deposits which are also subject to random withdrawals, justifying the need for a liquidity buffer. Funds only invest in short-term liabilities of banks and saitsfy a self-imposed liquidity ratio to serve daily random investor redemptions. The system can be shocked in two ways: A default of one or more banks on all their liabilities, or a change in investors' believes regarding the quality of one or more MMF portfolios. The model dynamics then are: Funds must engage in fire sales to serve large investor redemptions. Due to a lack of deep secondary markets for their portfolios, they do so at endogenously determined price discounts. Banks whose liabilities were subject of hese fire sales (simulating a lack of rolling over or reneewal of short-term funding from MMFs), will struggle to satisfy their liquidity and capital requirements, and ultimately will default. These defaults will cause problems for further banks and funds, depending on their portfolio overlaps. Finally, the underlying network structure is modelled as follows: For funds I make use of granular Form N-MFP portfolio data, providing linkages between funds and banks. Linkages within the banking sector are randomly generated, using a probability map (see e.g., Halaj and Kok (2013), based on the EBA large exposures data.
For the universe of Compustat firms, we construct firm-level intangible assets following the method of Peters and Taylor (2017) in order to answer the question of how the structural trend towards an intangible economy is affecting business cycle dynamics. In first tests, we confirm several empirical regularities along the intangibles dimension (such as low-intangible firms being larger, or high-intangible firms holding more cash). We further explore the puzzle of stagnating total factor productivity (TFP) measures. In our TFP construction we follow the methods of Olley & Pakes (1996) and show that for low-intangible firms, labor and physical capital are the main equally-important inputs, whereas for high-intangible firms the role of physical capital is, effectively, dropping to zero and intangible capital makes up almost 40% of the TFP for these firms.