Agree to Disagree: Measuring Hidden Dissents in FOMC Meetings
Available on SSRN, August 2023
Abstract Based on a record of dissents on FOMC votes and transcripts of the meetings from 1976 to 2017, we develop a deep learning model based on self-attention modules to create a measure of disagreement for each member in each meeting. While dissents are rare, we find that members often have reservations with the policy decision. The level of disagreement is mostly driven by current or predicted macroeconomic data at both the individual and meeting levels, while personal characteristics of the members matter only at the individual level. We also use our model to evaluate speeches made by members between meetings, and we find a weak correlation between the level of disagreement revealed in them and that of the following meeting. Finally, we find that the level of disagreement increases whenever monetary policy action is more aggressive.
The Impact of Stay-at-Home Orders on US Output: A Network Perspective
Available on SSRN, April 2020, Submitted
Abstract Under the stay-at-home orders issued by states, economic activities are reduced or put on hold by some states across the U.S. to control the spread of COVID-19. By combining several sources of data, we estimate the output loss due to such restrictions using a network approach. Based on our most conservative estimates, the measures as of April 15, 2020 reduce 26% of total US output per period, and about 43% of which is due to the input-output connections in the production network. Using a SIR model with an inter-state infection network, we also calculate the cost of reducing each infection to be approximately 150,000 dollars during the period of March 19 to April 15, 2020. Simulation results of various hypothetical stay-at-home orders show that the unit cost of infection reduction of the existing order is about 13% higher than the local minimum.
Do Connections Pay Off in the Bitcoin Market?
Published in Journal of Empirical Finance, February 2022 (with Kwok Ping Tsang)
Abstract This paper identifies the bitcoin investor network and studies the relationship between connections and returns. Using transaction data recorded in the bitcoin blockchain from 2015 to 2020, we reach three conclusions. First, connectedness is not strongly correlated with higher returns in the first four years. However, the correlation becomes strong and significant in 2019 and 2020. Second, returns also differ among those connected addresses. By dividing the connected addresses into ten decile groups based on their centrality, we find that the top 20% most-connected addresses earn higher returns than their peers during most of our sample period. Third, eigenvector centrality is more related to higher returns than degree centrality for the top 20% most-connected addresses, implying that the quality of connections may matter more than quantity among those highly connected addresses.
The Market for Bitcoin Transactions
Published in Journal of Internantional Financial Market, Institution & Money, January 2021 (with Kwok Ping Tsang)
Abstract Transaction fees in the bitcoin system work differently from those in conventional payment systems due to the design of the bitcoin mining algorithm. In particular, transaction fees and transaction volume in the bitcoin system increase whenever the network is congested, and our VAR results confirm that is indeed the case. To account for the empirical findings, we build a model where users and miners together determine transaction fees and transaction volume. Even though the mechanism of fluctuating transaction fees in bitcoin introduces an extra cost of uncertainty to users, a back-of-envelope calculation shows that the cost of using the bitcoin network for transactions is still smaller than the cost of using the current conventional payment system with a fixed transaction fee rate. However, this calculation may underestimate the cost due to the crowding-out effect on small transactions during the congested period.
Price dispersion in bitcoin exchanges
Published in Economics Letters, September 2020 (with Kwok Ping Tsang)
Abstract Bitcoin is traded in a number of exchanges, and there is a large and time-varying price dispersion among them. We identify the sources of price dispersion using a standard time-varying vector autoregression model with stochastic volatility, and we find that shocks to transaction fees and bitcoin price growth explain on average 20%, and sometimes more than 60%, of the variation of price dispersion.