Matthias C. Qian
Matthias is a Departmental Lecturer interested in a deep understanding of machine learning algorithms and their applications to solve real-world problems. He firmly believes that Artificial Intelligence will revolutionise the Economic Sciences and multiply what this discipline can contribute to society. He is a polymath in regard to machine learning techniques, studying and applying algorithms from Step-indicator Saturation to LSTMs and natural language processing techniques. He completed his MPhil and DPhil in Economics under the supervision of Professor Bent Nielsen. His undergraduate degree is from the Humboldt Universität in Berlin. He is an associate member at Nuffield College and the Oxford-Man Institute.
New working paper out on Entrepreneurship: Scaling Up Firms in Entrepreneurial Ecosystems: Fintech and Lawtech Ecosystems Compared.
UK Electric Vehicle and Battery Production Potential to 2040 (commissioned by the Faraday Institution), 10th June 2019; Oxford Uni news article here; The Guardian article here; the FT article here, the Telegraph article here.
A theoretical contribution of Matthias has been the study of the asymptotic theory of the Step-indicator Saturation estimator. This algorithm finds and models location shifts in time-series, which are changes in the mean of time-series. In the wake of significant disruptions of economic systems, such as the Global Financial Crisis or the Brexit transition process, algorithms that can handle structural breaks are more important than ever before. Matthias’ research gives this algorithm strengthened credibility, by proving a set of desirable theoretical properties that the algorithm exhibits.
His empirical work surround three economic markets, namely financial, property and labour markets. First, Matthias models global financial markets. He is particularly interested in unusual phenomena of these time-series. Phenomena include the breakdown of the efficient market hypothesis during commodity price crashes or the shift in the geography of price discovery during earning calls of dual-listed firms. Furthermore, he studies the outperformance of LSTM models in predicting financial volatility. He also established the co-integration relationship between deindustrialisation and the real interest rate. Overall, this project applies state-of-the-art machine learning to shed new light on traditional questions in finance.
Second, Matthias models the British residential property market. He combines property and individual level data to get a comprehensive understanding of the market. The project involves the merger of several large-scale datasets, including the public land registry and proprietary individual-level datasets. The project aims to predict which changes in property prices on a neighbourhood and property level and how policy and public investment decisions drive these price dynamics.
Third, Matthias models the dynamics of labour markets. As part of the ESRC project on English Law and AI, he uses online job vacancies to explain the effect of legislation on demand for labour in the United Kingdom. He also maps out the temporal change in the skill demand across vital service industries, including the legal, insurance and accounting industry. For the Faraday Institution, Matthias mapped out the adoption of Industry 4.0 techniques in battery plants across a range of geographies. The project aims to predict labour market trends using worker-level data.
We investigate the asymptotic properties of Step-indicator Saturation which is an algorithm to handle unmodelled location shifts in time series. We consider a stylized version of the algorithm that uses the split-half approach. We present asymptotic convergence and distribution results on the gauge of the algorithm which is the frequency of falsely retained step-indicators when the data generating process has no shifts. The proofs rely on empirical process results of temporal differences of residuals. Our results offer an asymptotic justification to use the gauge in choosing the tuning parameter of this statistical procedure.
Since the inception of the Bank of England the interest rates in the United Kingdom have never been as low as during the economic malaise that followed the Great Recession of 2007-09. This paper argues that deindustrialization, which we define as the declining output share of manufacturing, is a key driver for lower long-run real rates. We motivate the relationship using a general equilibrium model in the context of a trade shock which originates from the rise of China as a manufacturing power house. We test our hypothesis empirically using the catalyst method, which exploits the causal transmission of large exogenous shocks, to identify the unique direction of causation. From 1985 to 2017, we estimate that deindustrialization has reduced the British long-run real rates by 221 basis points.
Structural Change and Pairwise Co-integration in the Market for Crude Oil
We document an asset pricing anomaly in the presence of a temporary oversupply in the commodity markets. We show that a sufficiently prolonged period of low commodity prices results in a stable co-integrating relationship between the price of crude oil and the price of crude oil producing companies. The relationship is subject to structural breaks which are due to asset fire sales or extensions of credit lines. Step-indicator Saturation is used to estimate the cointegrating vector in the presence of the breaks. A simple mean-reverting trading strategy that exploits this co-integrating relationship generates substantial risk-adjusted returns.
We study the price discovery process of cross-listed and publicly listed firms on the B3 in Sao Paolo and the NYSE in New York. Earnings conference calls in English and Portuguese allow us to understand the evolution of the price discovery process when new information becomes available only to a subset of investors. We find that during English language conference calls, the NYSE has a higher information share in the price discovery processes than during Portuguese language conference calls.