Embedded Files
-- Working Papers --
Job Market Paper*
Transformers, Central Banker Speeches, Tail Stock Returns
I use transformers to construct central bank speech sentiment indices using financial text trained (expert) and not financial text trained (layman) models. I find that cross-sentence information is important for forecasting, and aggregating across positive and negative sentiment loses information. Empirically, sentiment measured by expert and aggregate indices affects stock returns. The share of tail stock returns increases (decreases) in response to negative Fed (Bank of Canada) speech sentiment. Negative Fed speech sentiment has a spillover effect with the same sign on the Canadian stock market, and leads changes in negative Canadian central banker speech sentiment. The relationship is driven by Fed Governor speeches. On trading days following Fed speeches, a long-only trading strategy based on return response to negative Fed speech sentiment could have outperformed the S&P 500 by 47% and 0.61 in terms of cumulative returns and Sharpe ratio respectively.
Krisztian_Gado_Job_Market_Paper.pdfWeather Disaster Attention and Daily REIT Returns: A Near Continuous Approach
This paper shows that daily REIT returns reflect prior day weather disaster attention. I find evidence for a time-varying relationship between previous day Weather Disaster Attention (WDA) and REIT holding period returns. The sign and persistence of the relationship varies over time. Within a given window, daily REIT returns exhibit a homogeneous, either mostly positive or negative, response to prior day WDA. I construct Weather Disaster Attention Factors (WDAFs), which periodically increase the performance of the Four Factor Model in terms of explaining variation in daily stock returns. I find no evidence of a risk premium associated with the WDAFs. Using long only positions, a WDAF based trading strategy consistently outperformed its relevant benchmarks in terms of realized as well as risk adjusted returns between January 2017 and February 2020. The implication is that markets are periodically inefficient with respect to information about natural disasters. I find no evidence that sticky expectations are the mechanism behind the WDAF based trading strategies’ profitability.
Krisztian_Gado_Climate_Attention_Paper.pdfAmazon and Online Competitor Prices: The Role of Shipping Costs in a Hybrid Platform Environment
With Raphael Schoenle & Stefan Thomas.
This analysis compares Amazon bestseller prices to competitor prices of identical products, sold both on Amazon Marketplace and in the entire online market space outside the Amazon platform. Amazon prices relative to all its competitors, or only those on Amazon marketplace, are generally widely dispersed, and both higher and lower. Crucially, the sign of the average percentage price difference tends to reverse depending on the inclusion or exclusion of shipping costs into prices. Notably, when including them into both Amazon and competitor prices, Amazon is on average less expensive. Comparing only prices of Amazon Marketplace sellers and outside online competitors, the median percentage difference is zero but substantially less dispersed. Shipping costs may also reverse the sign of this percentage price difference. These findings can help to assess platform-related pricing strategies under antitrust law.
Kris_G_Raphael_S_and_S_F_Amazon_Paper.pdfSociodemographic Disparities in Antibiotic-Resistant Outpatient Urine Cultures in a Boston Hospital, 2015-2020: A Cross-sectional Analysis
With Courtney W. Chan, Leo K. Westgard, Andrew Romasco, Shira Doron, and Maya L. Nadimpalli. Under Review at the IJEH.
Analyzing 1306 outpatient urine cultures at Tufts Medical Center from 2015-2020, we examine the influence of social determinants on antibiotic resistance (AR). We find that public insurance use increased the odds of resistance to first-line antibiotics, while living in a low-income neighborhood reduced the risk of multidrug-resistant (MDR) infections by 47%. A strong but non-significant trend suggested a higher likelihood of aminoglycoside resistance among non-English speakers. No significant associations were found between race or ethnicity and AR uropathogens.
-- Finished Work --
Beyond the Average: Patterns in UK Price Data at the Micro Level, with Lennart Brandt and Natalie Burr. Bank of England: Bank Underground, 17th of Jan, 2024. Available here.
Abstract: The Bank of England has a 2% annual inflation rate target in the ONS’ consumer prices index. But looking at its 700 item categories, we find that very few prices ever change by 2%. In fact, on a month-on-month basis, only about one fifth of prices change at all. Instead, we observe what economists call ‘sticky prices’: the price of an item will remain fixed for an extended amount of time and then adjust in one large step. We document the time-varying nature of stickiness by looking at the share of price changes and their distribution in the UK microdata. We find a visible discontinuity in price-setting in the first quarter of 2022, which has only partially unwound.
A Game Theoretic Analysis of Improvements to Tor's Resilience to Entry-Exit and End-to-End Attacks, with Jens Mache, Alexander Lotero and Lana Parezanin. In Proceedings of the 51st ACM Technical Symposium on Computer Science Education (SIGCSE ’20). Association for Computing Machinery, New York, NY, USA, 1425.
https://doi.org/10.1145/3328778.3372709
Abstract: Tor is among the most used overlay networks for anonymous communication. This anonymity can be undermined via entry-exit and end-to-end attacks. Using Game Theory, we analyze the viability of several methods for reducing Tor's vulnerability to such attacks. Entry-exit attacks rely upon controlling entry and exit nodes -- internal elements within the Tor Network, while end-to-end attacks utilize Autonomous Systems (internet service providers), elements outside of the Tor Network. Because both types of attacks rely on probability, we use Monte Carlo simulation and model the success probability maximizing strategies of adversaries. We analyze changes to Tor's node selection strategy that decrease the success probability of such attacks. Our goal is to support anonymity preserving systems against large Autonomous Systems providers and attackers with plenty of resources. We build upon previous work, but we also test eliminating the exit node bandwidth threshold and decreasing asymmetric routing to make compromising anonymity less likely. Given our results, we suggest the abandonment of the bandwidth threshold of exit nodes. Abandoning this threshold would not affect the bandwidth of the Tor network much, while it would impair an attacker's success probability significantly. We show that an attackers' success probability can be lowered by 23% over a year of usage. While prior to our changes (and assuming a large fraction of compromised nodes), anonymity is preserved ~62% of the time over a year, implementing our changes increases anonymity preservation to 85%.
Page updated
Google Sites
Report abuse