Research articles for the 2020-10-18

AAMDRL: Augmented Asset Management with Deep Reinforcement Learning
Eric Benhamou,David Saltiel,Sandrine Ungari,Abhishek Mukhopadhyay,Jamal Atif

Can an agent learn efficiently in a noisy and self adapting environment with sequential, non-stationary and non-homogeneous observations? Through trading bots, we illustrate how Deep Reinforcement Learning (DRL) can tackle this challenge. Our contributions are threefold: (i) the use of contextual information also referred to as augmented state in DRL, (ii) the impact of a one period lag between observations and actions that is more realistic for an asset management environment, (iii) the implementation of a new repetitive train test method called walk forward analysis, similar in spirit to cross validation for time series. Although our experiment is on trading bots, it can easily be translated to other bot environments that operate in sequential environment with regime changes and noisy data. Our experiment for an augmented asset manager interested in finding the best portfolio for hedging strategies shows that AAMDRL achieves superior returns and lower risk.

Ants, robots, humans: a self-organizing, complex systems modeling approach
Martin Jaraiz

Most of the grand challenges of humanity today involve complex agent-based systems, such as epidemiology, economics or ecology. However, remains as a pending task the challenge of identifying the general principles underlying their self-organizing capabilities. This article presents a novel modeling approach, capable to self-deploy both the system structure and the activities for goal-driven agents that can take appropriate actions to achieve their goals. Humans, robots, and animals are all endowed with this type of behavior. Self-organization is shown to emerge from the decisions of a common rational activity algorithm, based on the information of a system-specific goals dependency network. The unique self-deployment feature of this approach, that can also be applied to non-goal-driven agents, can boost considerably the range and depth of application of agent-based modeling.

Hybrid Modelling Approaches for Forecasting Energy Spot Prices in EPEC market
Tahir Miriyev,Alessandro Contu,Kevin Schafers,Ion Gabriel Ion

In this work we considered several hybrid modelling approaches for forecasting energy spot prices in EPEC market. Hybridization is performed through combining a Naive model, Fourier analysis, ARMA and GARCH models, a mean-reversion and jump-diffusion model, and Recurrent Neural Networks (RNN). Training data was given in terms of electricity prices for 2013-2014 years, and test data as a year of 2015.

Individual Heterogeneity and Cultural Attitudes in Credence Goods Provision
Johnny Tang

I study the heterogeneity of credence goods provision in taxi drivers taking detours in New York City. First, I document that there is significant detouring on average by drivers. Second, there is significant heterogeneity in cheating across individuals, yet each individual's propensity to take detours is stable: drivers who detour almost always detour, while those who do not detour almost never do. Drivers who take longer detours on each trip also take such trips more often. Third, cultural attitudes plausibly explain some of this heterogeneity in behavior across individuals.

Inversion-free Leontief inverse: statistical regularities in input-output analysis from partial information
Silvia Bartolucci,Fabio Caccioli,Francesco Caravelli,Pierpaolo Vivo

We present a baseline stochastic framework for assessing inter-sectorial relationships in a generic economy. We show that - irrespective of the specific features of the technology matrix for a given country or a particular year - the Leontief multipliers (and any upstreamness/downstreamness indicator computed from the Leontief inverse matrix) follow a universal pattern, which we characterize analytically. We formulate a universal benchmark to assess the structural inter-dependence of sectors in a generic economy. Several empirical results on World Input-Output Database (WIOD, 2013 Release) are presented that corroborate our findings.

KrigHedge: GP Surrogates for Delta Hedging
Mike Ludkovski,Yuri Saporito

We investigate a machine learning approach to option Greeks approximation based on Gaussian process (GP) surrogates. The method takes in noisily observed option prices, fits a nonparametric input-output map and then analytically differentiates the latter to obtain the various price sensitivities. Our motivation is to compute Greeks in cases where direct computation is expensive, such as in local volatility models, or can only ever be done approximately. We provide a detailed analysis of numerous aspects of GP surrogates, including choice of kernel family, simulation design, choice of trend function and impact of noise.

We further discuss the application to Delta hedging, including a new Lemma that relates quality of the Delta approximation to discrete-time hedging loss. Results are illustrated with two extensive case studies that consider estimation of Delta, Theta and Gamma and benchmark approximation quality and uncertainty quantification using a variety of statistical metrics. Among our key take-aways are the recommendation to use Matern kernels, the benefit of including virtual training points to capture boundary conditions, and the significant loss of fidelity when training on stock-path-based datasets.

Limits of Stress-Test Based Bank Regulation: Cues from the COVID-19 Crisis
Agarwal, Isha,Goel, Tirupam
Stress-tests can enhance welfare by providing complementary information about banks' risk exposures to regulators, which allows them to use capital surcharges to better align baseline regulation to individual banks. This paper provides suggestive evidence of inaccuracies in stress-testing using the COVID-19 crisis, and develops a model to study the attendant welfare consequences. We show that inaccuracies reduce welfare by causing excessive (insufficient) regulation of a less (more) risky bank, and by hampering banks' ex-ante incentives. Accuracy and the optimal surcharge have a non-linear relationship, and exhibit a phase shift -- for accuracy below a threshold, the optimal surcharge is zero.

Measuring the Effect of Unconventional Policies on Stock Market Volatility
Demetrio Lacava,Giampiero M. Gallo,Edoardo Otranto

As a response to the Great Recession, many central banks resorted to unconventional monetary policies, in the form of a balance sheet expansion. Our research aims at analyzing the impact of the ECB policies on stock market volatility in four Eurozone countries (France, Germany, Italy, and Spain) within the Multiplicative Error Model framework. We propose a model that allows us to quantify the part of market volatility depending directly on unconventional policies by distinguishing between the announcement effect and the implementation effect. While we observe an increase in volatility on announcement days, we find a negative implementation effect, which causes a remarkable reduction in volatility in the long term. A Model Confidence Set approach finds how the forecasting power of the proxy improves significantly after the policy announcement; a multi-step ahead forecasting exercise estimates the duration of the effect, and, by shocking the policy variable, we are able to quantify the reduction in volatility which is more marked for debt-troubled countries.

Parsimonious Quantile Regression of Financial Asset Tail Dynamics via Sequential Learning
Xing Yan,Weizhong Zhang,Lin Ma,Wei Liu,Qi Wu

We propose a parsimonious quantile regression framework to learn the dynamic tail behaviors of financial asset returns. Our model captures well both the time-varying characteristic and the asymmetrical heavy-tail property of financial time series. It combines the merits of a popular sequential neural network model, i.e., LSTM, with a novel parametric quantile function that we construct to represent the conditional distribution of asset returns. Our model also captures individually the serial dependences of higher moments, rather than just the volatility. Across a wide range of asset classes, the out-of-sample forecasts of conditional quantiles or VaR of our model outperform the GARCH family. Further, the proposed approach does not suffer from the issue of quantile crossing, nor does it expose to the ill-posedness comparing to the parametric probability density function approach.

The Impact of COVID-19 Pandemic on Bank Lending Around the World
Colak, Gonul,Öztekin, Özde
We evaluate the influence of the coronavirus pandemic on global bank lending and identify bank and country characteristics that amplify or weaken the effect of the disease outbreak on bank credit. Using a dataset comprising banks from 125 countries and a difference-in-difference methodology, we find that loan growth contracts considerably around the globe. This adverse effect depends on bank financial condition, bank market structure, country features including regulation and supervision of the banking systems, financial intermediary and debt market development, ease of access of corporate firms to debt capital, and the response of the public health sector to the crisis.

The measure of model risk in credit capital requirements
Roberto Baviera

Credit capital requirements in Internal Rating Based approaches require the calibration of two key parameters: the probability of default and the loss-given-default. This letter considers the uncertainty about these two parameters and models this uncertainty in an elementary way: it shows how this estimation risk can be computed and properly taken into account in regulatory capital.

We analyse two standard real datasets: one composed by all corporates rated by Moody's and one limited only to the speculative grade ones. We statistically test model hypotheses on both marginal distributions and parameter dependency. We compute the estimation risk impact and observe that parameter dependency raises substantially the tail risk in capital requirements. The results are striking with a required increase in regulatory capital in the range $38\%$-$66\%$.