Consortium for Data Analytics in Risk

Lisa Goldberg, co-Director, CDAR

“Machine learning for inclusive finance: thoughts and practice”
Yuan (Alan) Qi, Ant Financial

Talk Abstract: Machine learning is a critical driving force for inclusive finance. In this talk, I will discuss some important challenges we face in addressing inclusive finance problems, and present several machine learning technologies we have developed to overcome these challenges and their applications in microlending, risk control, insurance claiming process, marketing and customer service.

Talk Abstract: This paper investigates the divergence of environmental, social, and governance (ESG) ratings. First, the paper documents the disagreement between the ESG ratings of five prominent rating agencies. The paper proceeds to trace the disagreement to the most granular level of ESG categories that is available and decomposes the overall divergence into three sources: Scope divergence related to the selection of different sets of categories, measurement divergence related to different assessment of ESG categories, and weight divergence related to the relative importance of categories in the computation of the aggregate ESG score. We find that measurement divergence explains more than 50 percent of the overall divergence. Scope and weight divergence together are slightly less important. In addition, we detect a rater effect, i.e., the rating agencies’ assessment in individual categories seems to be influenced by their view of the analyzed company as a whole. The results allow investors, companies, and researchers to understand why ESG ratings differ.

Talk Abstract: Understanding the social and economic consequences of climate change is fundamentally a risk analysis, since climate change alters the underlying probabilities that different types of natural events (e.g. hurricanes) occur. I will describe many of the advances we have made in applying risk analysis to evaluate the potential impacts of climate change on society and the economy, from affecting civil wars and human health to slowing GDP growth and accelerating asset depreciation. I will discuss how these analyses affect what we understand about the value of investments that mitigate climate change or help us adapt to its effects.

Talk Abstract: Traditionally, practitioners have believed that stock return factor models are not stable over histories longer than a medium horizon of T_M = 250 or T_M = 500 trading days (one or two years). We first show that Principal Components Analysis (PCA) consistently estimates factor loadings under essentially arbitrary variable volatility of factors. Then, we demonstrate empirically that the use of long histories such as T_L = 1750 trading days, or seven years, improves the performance of optimized portfolios, compared to using a medium history of T_M days, even when corrected for excess dispersion as proposed by Goldberg, Papanicalaou and Shkolnik (GPS). We find that the factor loadings of individual stocks and portfolios on long-horizon PCA factors are, in fact, relatively stable.  We estimate factor loadings with a long history T_L and estimate the current covariance matrix of the factors with a short data history (half-life T_S = 40 days).

Talk Abstract: This paper combines dimensional analysis, leverage neutrality, and a principle of market microstructure invariance to derive scaling laws expressing transaction costs functions, bid-ask spreads, bet sizes, number of bets, and other financial variables in terms of dollar trading volume and volatility. The scaling laws are illustrated using data on bid-ask spreads and number of trades for Russian and U.S. stocks. These scaling laws provide practical metrics for risk managers and traders; scientific benchmarks for evaluating controversial issues related to high frequency trading, market crashes, and liquidity measurement; and guidelines for designing policies in the aftermath of financial crisis.

Moderator:

Jeffrey R. Bohn, Swiss Re

Panelists:

Mallik Tatipamula, Ericsson North America

Nicole Hu, One Concern

Stuart Evans, CMU-Emirates iLab

Abstract:

The collection of widely available artificial intelligence (AI) & machine learning (ML) tools has become a catalyst for new approaches to transform processes ranging from image recognition to self-driving vehicles. Some of these AI & ML tools have revolutionized the way advertising targets consumers. Despite these successes, AI & ML tools have yet to materially transform most large corporations. Regardless, most large firms continue to invest heavily in these areas. These investments tend to be skewed toward algorithm development with insufficient investment in data collection & curation and organizational education & design. Little investment focuses on how best to make productive use of insights arising from these tools. This lack of focus on end-to-end implementations of these AI & ML tools is one hypothesis for why these new tools have had so little impact on corporate profitability. Another hypothesis for failure relates to mis-application of particular approaches/algorithms to use cases.

This panel will explore these hypotheses and discuss the state-of-the-art in AI & ML for large firms with a focus on what can be done to realize value from investments in workflow areas where these tools should produce more value.

Areas where AI & ML tools should be more productive (but are not—despite considerable investment):

1. Predicting time series for business, economic, and financial analyses
2. Identifying fraud
3. Distilling useful information from multiple sources of heterogeneous sources of unstructured data
4. Assessing risk
5. Managing capital
6. Developing risk & business scenarios

Lisa Goldberg, CDAR