Understanding financial stability: climate-related considerations and financial markets operations

This dissertation is a collection of four empirical studies in the fields of financial economics.
The first and second studies relate to the literature of climate economics and contribute to the policy discussions on climate-related transition risk. The last two studies build upon the literature of market microstructure and contribute to the policy discussions on speed in financial markets. The first study of this dissertation investigates whether and how climate-related transition metrics of European large corporate firms relate to the credit risk of these firms implied by credit default swap for different time horizons. Based on an empirical analysis of firm-specific historical data, I find that firms with higher Scope 1 GHG emissions have higher CDS-implied credit risk. This relationship is reflected even by 30-year CDS, particularly after 2015 when a shift in market awareness of transition risk occurred. Albeit the European CDS market is already pricing to some small extent the effect of emissions at different time horizons, other material climate-related transition metrics do not yet reflected. In the second study, we examine how climate-related transition metrics relate to firms’ credit ratings of corporate firms in advanced economies such as Europe and the USA. We find that high emissions tend to be associated with worse credit ratings. Yet firms that disclose emissions and a forward-looking commitment to cut emissions have lower credit risk, with the effect tending to be stronger for more ambitious targets. We also find that after the Paris agreement in 2015, European firms most exposed to climate transition risks saw their ratings deteriorate. The effect is larger for European than US firms, probably reflecting differential expectations around climate policy. The third study investigates how resilient are modern trading venues in a high-frequency environment with cross-venue fragmented order flow. We build a unique cross-venue dataset with millisecond resolution, covering two major competing stock exchanges, London Stock Exchange and Chi-X. Employing a Hawkes process methodology, we find that the average time for the stock market to return to normal after a shock is below known human response times, which is circa 600 milliseconds, suggesting that a substantial amount of stock market activity is run by trading bots responses. In the fourth study, we turn our attention to high frequency versus low frequency market data. We explore a set of neural network machine learning models on news and financial data to predict shock events in high-frequency and low-frequency market data. We find that the market movement in response to a piece of news may be instantaneous or, most likely, cumulative over time.