Title: "Implementing an operational process for scoring first notifications of loss for claims fraud."
At the first notification of an insurance claim, it is useful for insurance companies to estimate the probability that fraud is involved. This short talk looks at some aspects of implementing an operational process for creating these scores. Time permitting, topics that will be covered include the basic supervised machine learning techniques used; shortcomings of the basic techniques; practical challenges; skillsets required.
Hans-Jörg von Mettenheim.
Title: "Applied Artificial Intelligence for Quantitative Trading"
The buzzwords "artificial intelligence", "deep learning", and "big data" have taken over the algorithmic and quantitative trading world by storm over the past few years. This talk intends to shed some light on the different domains of machine learning applications in quantitative trading: what works well, what works OK, what is dubious? Often, a combination of "advanced" techniques with more "traditional" techniques in the shape of a hybrid model lead to promising results. Therefore, the presentation will also include an example of this type of mixed models.
Title: "Macroeconomics and Machine Learning"
The fields of macroeconomics and applied economics inherit from a long tradition of econometric models. Traditional econometric and parametric models are thus still used today by economics and finance practitioners. Machine learning, big data and AI even have a bad reputation: too complex, black box, not useful for macroeconomics, etc. But things are gradually changing, and in recent years a number of practitioners and academic researchers have paved the way for practical applications of these new quantitative methods to economics and finance. During this presentation, I will show illustrations of the use of ML/AI techniques on country risk analysis, on US business cycle analysis, and on text mining techniques applied to publications on emerging markets.
Title: "Interpretability and Explainability in Machine Learning: A Sensitivity Analysis Viewpoint"
Increasing research efforts are devoted to augment the interpretability of machine findings. When complex architectures are used, analysts are, in fact, exposed to the black-box effect. This seminar will review several methods used both in the machine learning and in the simulation community to make the black box more transparent. We shall discuss tools such as partial dependence functions, layerwise relevance propagation, as well as present several local and global sensitivity analysis methods, also proposing new tools and new findings on popular machine learning tools.
Manuel Nunes and Frank McGroarty
Title: "LSTM-LagLasso for bond yield forecasting: Peeping into the long short-term memory network's black box"
Modern fixed income asset management requires the development of intelligent systems, which involve the use of state-of-the-art machine learning models and appropriate methodologies. We conduct the first study of bond yield forecasting using long short-term memory (LSTM) networks, validating its potential and identifying its memory advantage. Specifically, we model the 10-year bond yield using univariate LSTMs with three input sequences and five forecasting horizons. We compare those with multilayer perceptrons (MLP), univariate and with the most relevant features. To demystify the notion of black box associated with LSTMs, we conduct the first internal study of the model. To this end, we calculate the LSTM signals through time, at selected locations in the memory cell, using sequence-to-sequence architectures, uni- and multi-variate. We then proceed to explain the states’ signals using exogenous information, for what we develop the LSTM-LagLasso methodology. The results show that the univariate LSTM model with additional memory is capable of achieving similar results as the multivariate MLP using macroeconomic and market information. Moreover, the direct comparison leads to results with LSTMs that are similar or better with lower standard deviations. Furthermore, shorter forecasting horizons require smaller input sequences and vice-versa. The most remarkable property found consistently in the LSTM signals, is the activation/deactivation of units through time, and the specialisation of units by yield range or feature. Those signals are complex but can be explained by exogenous macroeconomic and market variables. Additionally, some of the relevant features identified via LSTM-LagLasso are not commonly used in forecasting models, and lags are important.