|
on Big Data |
| By: | Nils K\"obis; Luca Mossink |
| Abstract: | The release of openly available, robust text generation algorithms has spurred much public attention and debate, due to algorithm's purported ability to generate human-like text across various domains. Yet, empirical evidence using incentivized tasks to assess human behavioral reactions to such algorithms is lacking. We conducted two experiments assessing behavioral reactions to the state-of-the-art Natural Language Generation algorithm GPT-2 (Ntotal = 830). Using the identical starting lines of human poems, GPT-2 produced samples of multiple algorithmically-generated poems. From these samples, either a random poem was chosen (Human-out-of-the-loop) or the best one was selected (Human-in-the-loop) and in turn matched with a human written poem. Taking part in a new incentivized version of the Turing Test, participants failed to reliably detect the algorithmically-generated poems in the human-in-the-loop treatment, yet succeeded in the Human-out-of-the-loop treatment. Further, the results reveal a general aversion towards algorithmic poetry, independent on whether participants were informed about the algorithmic origin of the poem (Transparency) or not (Opacity). We discuss what these results convey about the performance of NLG algorithms to produce human-like text and propose methodologies to study such learning algorithms in experimental settings. |
| Date: | 2020–05 |
| URL: | http://d.repec.org/n?u=RePEc:arx:papers:2005.09980&r=all |
| By: | Marcin Chlebus (Faculty of Economic Sciences, University of Warsaw); Maciej Stefan Świtała (Faculty of Economic Sciences, University of Warsaw) |
| Abstract: | The paper takes into consideration the broad idea of topic modelling and its application. The aim of the research was to identify mutual tendencies in econometric and machine learning abstracts. Different topic models were compared in terms of their performance and interpretability. The former was measured with a newly introduced approach. Summaries collected from esteemed journals were analysed with LSA, LDA and CTM algorithms. The obtained results enable finding similar trends in both corpora. Probabilistic models – LDA and CTM – outperform the semantic alternative – LSA. It appears that econometrics and machine learning are fields that consider problems being rather homogenous at the level of concept. However, they differ in terms of used tools and dominance in particular areas. |
| Keywords: | abstracts, comparison, interpretability, tendencies, topics |
| JEL: | A12 C18 C38 C52 C61 |
| Date: | 2020 |
| URL: | http://d.repec.org/n?u=RePEc:war:wpaper:2020-16&r=all |
| By: | Takanobu Mizuta |
| Abstract: | Who should be charged with responsibility for an artificial intelligence performing market manipulation have been discussed. In this study, I constructed an artificial intelligence using a genetic algorithm that learns in an artificial market simulation, and investigated whether the artificial intelligence discovers market manipulation through learning with an artificial market simulation despite a builder of artificial intelligence has no intention of market manipulation. As a result, the artificial intelligence discovered market manipulation as an optimal investment strategy. This result suggests necessity of regulation, such as obligating builders of artificial intelligence to prevent artificial intelligence from performing market manipulation. |
| Date: | 2020–05 |
| URL: | http://d.repec.org/n?u=RePEc:arx:papers:2005.10488&r=all |
| By: | Alain Naef (University of California, Berkeley) |
| Abstract: | Few studies on foreign exchange intervention convincingly address the causal effect of intervention on exchange rates. By using a narrative approach, I address a major issue in the literature: the endogeneity of intraday news which influence the exchange rate alongside central bank operations. Some studies find that interventions work in up to 80% of cases. Yet, by accounting for intraday market moving news, I find that in adverse conditions, the Bank of England managed to influence the exchange rate only in 8% of cases. I use both machine learning and human assessment to confirm the validity of the narrative approach. |
| Keywords: | intervention, foreign exchange, natural language processing, central bank, Bank of England. |
| JEL: | F31 E5 N14 N24 |
| Date: | 2020–06 |
| URL: | http://d.repec.org/n?u=RePEc:hes:wpaper:0188&r=all |
| By: | Franklin Allen; Xian Gu; Julapa Jagtiani |
| Abstract: | The intersection of finance and technology, known as fintech, has resulted in the dramatic growth of innovations and has changed the entire financial landscape. While fintech has a critical role to play in democratizing credit access to the unbanked and thin-file consumers around the globe, those consumers who are currently well served also turn to fintech for faster services and greater transparency. Fintech, particularly the blockchain, has the potential to be disruptive to financial systems and intermediation. Our aim in this paper is to provide a comprehensive fintech literature survey with relevant research studies and policy discussion around the various aspects of fintech. The topics include marketplace and peer-to-peer lending, credit scoring, alternative data, distributed ledger technologies, blockchain, smart contracts, cryptocurrencies and initial coin offerings, central bank digital currency, robo-advising, quantitative investment and trading strategies, cybersecurity, identity theft, cloud computing, use of big data and artificial intelligence and machine learning, identity and fraud detection, anti-money laundering, Know Your Customers, natural language processing, regtech, insuretech, sandboxes, and fintech regulations. |
| Keywords: | fintech; marketplace lending; P2P; alternative data; DLT; blockchain; robo advisor; regtech; insuretech; cryptocurrencies; ICOs; CBDC; cloud computing; AML; KYC; NLP; fintech regulations |
| JEL: | G21 G28 G18 L21 |
| Date: | 2020–05–28 |
| URL: | http://d.repec.org/n?u=RePEc:fip:fedpwp:88120&r=all |
| By: | Jie Fang; Jianwu Lin |
| Abstract: | One of the major characteristics of financial time series is that they contain a large amount of non-stationary noise, which is challenging for deep neural networks. People normally use various features to address this problem. However, the performance of these features depends on the choice of hyper-parameters. In this paper, we propose to use neural networks to represent these indicators and train a large network constructed of smaller networks as feature layers to fine-tune the prior knowledge represented by the indicators. During back propagation, prior knowledge is transferred from human logic to machine logic via gradient descent. Prior knowledge is the deep belief of neural network and teaches the network to not be affected by non-stationary noise. Moreover, co-distillation is applied to distill the structure into a much smaller size to reduce redundant features and the risk of overfitting. In addition, the decisions of the smaller networks in terms of gradient descent are more robust and cautious than those of large networks. In numerical experiments, we find that our algorithm is faster and more accurate than traditional methods on real financial datasets. We also conduct experiments to verify and comprehend the method. |
| Date: | 2020–06 |
| URL: | http://d.repec.org/n?u=RePEc:arx:papers:2006.09247&r=all |
| By: | Matteo Gambara; Josef Teichmann |
| Abstract: | Consistent Recalibration models (CRC) have been introduced to capture in necessary generality the dynamic features of term structures of derivatives' prices. Several approaches have been suggested to tackle this problem, but all of them, including CRC models, suffered from numerical intractabilities mainly due to the presence of complicated drift terms or consistency conditions. We overcome this problem by machine learning techniques, which allow to store the crucial drift term's information in neural network type functions. This yields first time dynamic term structure models which can be efficiently simulated. |
| Date: | 2020–06 |
| URL: | http://d.repec.org/n?u=RePEc:arx:papers:2006.09455&r=all |
| By: | Jennifer L. Castle (Dept of Economics, Institute for New Economic Thinking at the Oxford Martin School and Magdalen College, University of Oxford); Jurgen A. Doornik (Dept of Economics, Institute for New Economic Thinking at the Oxford Martin School and Climate Econometrics, Nuffield College, University of Oxford); David F. Hendry (Dept of Economics, Institute for New Economic Thinking at the Oxford Martin School and Climate Econometrics, Nuffield College, University of Oxford) |
| Abstract: | We have been publishing real-time forecasts of confirmed cases and deaths for COVID-19 online at www.doornik.com/COVID-19 from mid-March 2020. These forecasts are short-term statistical extrapolations of past and current data. They assume that the underlying trend is informative of short term developments, without requiring other assumptions of how the SARS-CoV-2 virus is spreading, or whether preventative policies are effective. As such they are complementary to forecasts from epidemiological models. The forecasts are based on extracting trends from windows of the data, applying machine learning, and then computing forecasts by applying some constraints to this flexible extracted trend. The methods have previously been applied to various other time series data and have performed well. They are also effective in this setting, providing better forecasts than some epidemiological models. |
| Keywords: | Autometrics; Cardt; COVID-19; Epidemiology; Forecasting; Forecast averaging; Machine learning; Smoothing; Trend Indicator Saturation. |
| Date: | 2020–04–27 |
| URL: | http://d.repec.org/n?u=RePEc:nuf:econwp:2006&r=all |
| By: | Laura Leal; Mathieu Lauri\`ere; Charles-Albert Lehalle |
| Abstract: | We use a deep neural network to generate controllers for optimal trading on high frequency data. For the first time, a neural network learns the mapping between the preferences of the trader, i.e. risk aversion parameters, and the optimal controls. An important challenge in learning this mapping is that in intraday trading, trader's actions influence price dynamics in closed loop via the market impact. The exploration--exploitation tradeoff generated by the efficient execution is addressed by tuning the trader's preferences to ensure long enough trajectories are produced during the learning phase. The issue of scarcity of financial data is solved by transfer learning: the neural network is first trained on trajectories generated thanks to a Monte-Carlo scheme, leading to a good initialization before training on historical trajectories. Moreover, to answer to genuine requests of financial regulators on the explainability of machine learning generated controls, we project the obtained "blackbox controls" on the space usually spanned by the closed-form solution of the stylized optimal trading problem, leading to a transparent structure. For more realistic loss functions that have no closed-form solution, we show that the average distance between the generated controls and their explainable version remains small. This opens the door to the acceptance of ML-generated controls by financial regulators. |
| Date: | 2020–06 |
| URL: | http://d.repec.org/n?u=RePEc:arx:papers:2006.09611&r=all |
| By: | Dhagash Mehta; Dhruv Desai; Jithin Pradeep |
| Abstract: | Given the surge in popularity of mutual funds (including exchange-traded funds (ETFs)) as a diversified financial investment, a vast variety of mutual funds from various investment management firms and diversification strategies have become available in the market. Identifying similar mutual funds among such a wide landscape of mutual funds has become more important than ever because of many applications ranging from sales and marketing to portfolio replication, portfolio diversification and tax loss harvesting. The current best method is data-vendor provided categorization which usually relies on curation by human experts with the help of available data. In this work, we establish that an industry wide well-regarded categorization system is learnable using machine learning and largely reproducible, and in turn constructing a truly data-driven categorization. We discuss the intellectual challenges in learning this man-made system, our results and their implications. |
| Date: | 2020–05 |
| URL: | http://d.repec.org/n?u=RePEc:arx:papers:2006.00123&r=all |
| By: | Sumit Agarwal; John Grigsby; Ali Hortaçsu; Gregor Matvos; Amit Seru; Vincent Yao |
| Abstract: | We study the interaction of search and application approval in credit markets. We combine a unique dataset, which details search behavior for a large sample of mortgage borrowers, with loan application and rejection decisions. Our data reveal substantial dispersion in mortgage rates and search intensity, conditional on observables. However, in contrast to predictions of standard search models, we find a novel non-monotonic relationship between search and realized prices: borrowers, who search a lot, obtain more expensive mortgages than borrowers' with less frequent search. The evidence suggests that this occurs because lenders screen borrowers' creditworthiness, rejecting unworthy borrowers, which differentiates consumer credit markets from other search markets. Based on these insights, we build a model that combines search and screening in presence of asymmetric information. Risky borrowers internalize the probability that their application is rejected, and behave as if they had higher search costs. The model rationalizes the relationship between search, interest rates, defaults, and application rejections, and highlights the tight link between credit standards and pricing. We estimate the parameters of the model and study several counterfactuals. The model suggests that overpayment may be a poor proxy for consumer unsophistication since it partly represents rational search in presence of rejections. Moreover, the development of improved screening technologies from AI and big data (i.e., fintech lending) could endogenously lead to more severe adverse selection in credit markets. Finally, place based policies, such as the Community Reinvestment Act, may affect equilibrium prices through endogenous search responses rather than increased credit risk. |
| JEL: | G21 L00 |
| Date: | 2020–06 |
| URL: | http://d.repec.org/n?u=RePEc:nbr:nberwo:27341&r=all |
| By: | Stetter, Christian; Menning, Philipp; Sauer, Johannes |
| Abstract: | Legislators in the EU have long been concerned with the environmental impact of farming activities. As a means to mitigate adverse ecological effects and foster desirable ecosystem services in agriculture, the EU introduced so-called agri-environment schemes (AES). This study suggests a machine learning method based on generalized random forests (GRF) for assessing the environmental effectiveness of such agri-environment payment schemes at the farm-level. We exploit a set of more than 130 contextual predictors to assess the individual impact of participating in agri-environment schemes in the EU. Results from our empirical application for Southeast Germany suggest the existence of heterogeneous impacts of environmental subsidies on mineral fertiliser quantities, greenhouse gas emissions and crop diversity. Individual treatment effects largely differ from traditionally used average treatment effects, thus indicating the importance of considering the farming context in agricultural policy evaluation. Furthermore, we provide important insights into the optimal targeting of agrienvironment schemes for maximising the environmental efficacy of existing policies. |
| Keywords: | Agricultural and Food Policy, Environmental Economics and Policy, Farm Management |
| Date: | 2020–04 |
| URL: | http://d.repec.org/n?u=RePEc:ags:aesc20:303699&r=all |
| By: | codagnone, cristiano; Bogliacino, Francesco (Universidad Nacional de Colombia); Gómez, Camilo Ernesto (Centro de Investigaciones para el Desarrollo); Charris, Rafael Alberto (Universidad Nacional de Colombia); Montealegre, Felipe (Universidad Nacional de Colombia); Liva, Giovanni; Villanueva, Francisco Lupiañez; Folkvord, F.; Veltri, Giuseppe Alessandro Prof (University of Trento) |
| Abstract: | Currently, many different countries are under lockdown or extreme social distancing measures to control the spread of COVID-19. The potentially far-reaching side effects of these measures have not yet been fully understood. In this study we analyse the results of a multi-country survey conducted in Italy (N=3,504), Spain (N=3,524) and the United Kingdom (N=3,523), with two separate analyses. In the first analysis, we examine the elicitation of citizens’ concerns over the downplaying of the economic consequences of the lockdown during the COVID-19 pandemic. We control for Social Desirability Bias through a list experiment included in the survey. In the second analysis, we examine the data from the same survey to estimate the consequences of the economic lockdown in terms of mental health, by predicting the level of stress, anxiety and depression associated with being economically vulnerable and having been affected by a negative economic shock. To accomplish this, we have used a prediction algorithm based on machine learning techniques. To quantify the size of this affected population, we compare its magnitude with the number of people affected by COVID-19 using measures of susceptibility, vulnerability and behavioural change collected in the same questionnaire. We find that the concern for the economy and for “the way out” of the lockdown is diffuse and there is evidence of minor underreporting. Additionally, we estimate that around 42.8% of the populations in the three countries are at high risk of stress, anxiety and depression, based on their level of economic vulnerability and their exposure to a negative economic shock. Therefore, it can be concluded that the lockdown and extreme social distancing in the three countries has had an enormous impact on individuals’ mental health and this should be taken into account for future decisions made on regulations concerning the pandemic. |
| Date: | 2020–05–30 |
| URL: | http://d.repec.org/n?u=RePEc:osf:socarx:x9m36&r=all |
| By: | Karolina Sowinska; Pranava Madhyastha |
| Abstract: | Public opinion influences events, especially related to stock market movement, in which a subtle hint can influence the local outcome of the market. In this paper, we present a dataset that allows for company-level analysis of tweet based impact on one-, two-, three-, and seven-day stock returns. Our dataset consists of 862, 231 labelled instances from twitter in English, we also release a cleaned subset of 85, 176 labelled instances to the community. We also provide baselines using standard machine learning algorithms and a multi-view learning based approach that makes use of different types of features. Our dataset, scripts and models are publicly available at: https://github.com/ImperialNLP/stockretu rnpred. |
| Date: | 2020–06 |
| URL: | http://d.repec.org/n?u=RePEc:arx:papers:2006.09723&r=all |
| By: | Yun-Cheng Tsai; Chun-Chieh Wang |
| Abstract: | Reinforcement learning can interact with the environment and is suitable for applications in decision control systems. Therefore, we used the reinforcement learning method to establish a foreign exchange transaction, avoiding the long-standing problem of unstable trends in deep learning predictions. In the system design, we optimized the Sure-Fire statistical arbitrage policy, set three different actions, encoded the continuous price over a period of time into a heat-map view of the Gramian Angular Field (GAF) and compared the Deep Q Learning (DQN) and Proximal Policy Optimization (PPO) algorithms. To test feasibility, we analyzed three currency pairs, namely EUR/USD, GBP/USD, and AUD/USD. We trained the data in units of four hours from 1 August 2018 to 30 November 2018 and tested model performance using data between 1 December 2018 and 31 December 2018. The test results of the various models indicated that favorable investment performance was achieved as long as the model was able to handle complex and random processes and the state was able to describe the environment, validating the feasibility of reinforcement learning in the development of trading strategies. |
| Date: | 2019–08 |
| URL: | http://d.repec.org/n?u=RePEc:arx:papers:1908.08036&r=all |
| By: | Carvalho, V; Garcia, Juan R.; Hansen, S.; Ortiz, A.; Rodrigo, T.; More, J. V. R. |
| Abstract: | We exploit high-frequency/high-resolution transaction data from BBVA, the second-largest bank in Spain, to analyse the dynamics of expenditure in Spain during the ongoing COVID-19 pandemic. Our main dataset consists of the universe of BBVA-mediated sales transactions from both credit cards and point-of-sales terminals, and totals 1.4 billion individual transactions since 2019. This dataset provides a unique opportunity to study the impact of the ongoing crisis in Spain—and the policies put in place to control it—on a daily basis. We find little shift in expenditure prior to the national lockdown, but then immediate, very large, and sustained expenditure reductions thereafter. Transaction metadata also allows us to study variation in these reductions across geography, sectors, and mode of sale (e.g. online/offline). We conclude that transaction data captures many salient patterns in how an economy reacts to shocks in real time, which makes its potential value to policy makers and researchers high. |
| Date: | 2020–04–14 |
| URL: | http://d.repec.org/n?u=RePEc:cam:camdae:2030&r=all |
| By: | Carvalho, Vasco M; Hansen, Stephen; Ortiz, Álvaro; Ramón García, Juan; Rodrigo, Tomasa; Rodriguez Mora, Sevi; Ruiz, José |
| Abstract: | We exploit high-frequency/high-resolution transaction data from BBVA, the second-largest bank in Spain, to analyse the dynamics of expenditure in Spain during the ongoing COVID-19 pandemic. Our main dataset consists of the universe of BBVA-mediated sales transactions from both credit cards and point-of-sales terminals, and totals 1.4 billion individual transactions since 2019. This dataset provides a unique opportunity to study the impact of the ongoing crisis in Spain--and the policies put in place to control it--on a daily basis. We find little shift in expenditure prior to the national lockdown, but then immediate, very large, and sustained expenditure reductions thereafter. Transaction metadata also allows us to study variation in these reductions across geography, sectors, and mode of sale (e.g. online/offline). We conclude that transaction data captures many salient patterns in how an economy reacts to shocks in real time, which makes its potential value to policy makers and researchers high. |
| Date: | 2020–04 |
| URL: | http://d.repec.org/n?u=RePEc:cpr:ceprdp:14642&r=all |
| By: | Gallego, J; Prem, M; Vargas, J. F |
| Abstract: | The public health crisis caused by the COVID-19 pandemic, coupled with the subsequent economic emergency and social turmoil, has pushed governments to substantially and swiftly increase spending. Because of the pressing nature of the crisis, public procurement rules and procedures have been relaxed in many places in order to expedite transactions. However, this may also create opportunities for corruption. Using contract-level information on public spending from Colombia’s e-procurement platform, and a differencein-differences identification strategy, we find that municipalities classified by a machine learning algorithm as traditionally more prone to corruption react to the pandemic-led spending surge by using a larger proportion of discretionary non-competitive contracts and increasing their average value. This is especially so in the case of contracts to procure crisisrelated goods and services. Our evidence suggests that large negative shocks that require fast and massive spending may increase corruption, thus at least partially offsetting the mitigating effects of this fiscal instrument. |
| Keywords: | Corruption, COVID-19, Public procurement, Machine learning |
| JEL: | H57 H75 D73 I18 |
| Date: | 2020–05–22 |
| URL: | http://d.repec.org/n?u=RePEc:col:000092:018178&r=all |
| By: | Thuy D. Nguyen; Sumedha Gupta; Martin Andersen; Ana Bento; Kosali I. Simon; Coady Wing |
| Abstract: | We study the effect of state reopening policies on multiple measures of human mobility and interaction during the COVID-19 epidemic. We document the timing and gradualness of the reopening plans that are occurring in different states, and we use data on multiple measures based on cell device signals to assess behavior during the reopening phase. Our empirical analysis suggests four main results. First, we document that after a substantial decline during the lock down period, there as been a clear increase in mobility levels in most states since mid-April. The resurgence of mobility is small relative to the decline since mid-March, but it is observable across a broad range of different cell device based metrics. Second, the size of the increase in mobility across counties is strongly associated with temperature and precipitation patterns. Third, event study analysis finds that by four days after reopening, the causal effect of policy on most measures of mobility is an increase of about 4% to 8%. Demonstrating the importance of using multiple measures of human mobility, the full range of point estimates spans 1% to 22%, without substantial pre-trend concerns. For example, there are larger effects on the variety of venues people visit than on the fraction leaving their homes. Fourth, the largest increase in mobility from reopening seems to occur in states that were late adopters of closure measures, suggesting that closure policies may have represented more of a binding constraint in those states. |
| JEL: | I1 |
| Date: | 2020–05 |
| URL: | http://d.repec.org/n?u=RePEc:nbr:nberwo:27235&r=all |
| By: | Jun-Hao Chen; Samuel Yen-Chi Chen; Yun-Cheng Tsai; Chih-Shiang Shur |
| Abstract: | Deep learning (DL) has been applied extensively in a wide range of fields. However, it has been shown that DL models are susceptible to a certain kinds of perturbations called \emph{adversarial attacks}. To fully unlock the power of DL in critical fields such as financial trading, it is necessary to address such issues. In this paper, we present a method of constructing perturbed examples and use these examples to boost the robustness of the model. Our algorithm increases the stability of DL models for candlestick classification with respect to perturbations in the input data. |
| Date: | 2020–05 |
| URL: | http://d.repec.org/n?u=RePEc:arx:papers:2006.03686&r=all |
| By: | Fred Espen Benth; Nils Detering; Silvia Lavagnini |
| Abstract: | We price European-style options written on forward contracts in a commodity market, which we model with a state-dependent infinite-dimensional Heath-Jarrow-Morton (HJM) approach. We introduce a new class of volatility operators which map the square integrable noise into the Filipovi\'{c} space of forward curves, and we specify a deterministic parametrized version of it. For calibration purposes, we train a neural network to approximate the option price as a function of the model parameters. We then use it to calibrate the HJM parameters starting from (simulated) option market data. Finally we introduce a new loss function that takes into account bid and ask prices and offers a solution to calibration in illiquid markets. A key issue discovered is that the trained neural network might be non-injective, which could potentially lead to poor accuracy in calibrating the forward curve parameters, even when showing a high degree of accuracy in recovering the prices. This reveals that the original meaning of the parameters gets somehow lost in the approximation. |
| Date: | 2020–06 |
| URL: | http://d.repec.org/n?u=RePEc:arx:papers:2006.01911&r=all |
| By: | Malo Huard (LMO - Laboratoire de Mathématiques d'Orsay - Université Paris-Saclay - CNRS - Centre National de la Recherche Scientifique, Milvue); Rémy Garnier (Cdiscount); Gilles Stoltz (LMO - Laboratoire de Mathématiques d'Orsay - Université Paris-Saclay - CNRS - Centre National de la Recherche Scientifique, HEC Paris - Ecole des Hautes Etudes Commerciales, CELESTE - Statistique mathématique et apprentissage - Inria Saclay - Ile de France - Inria - Institut National de Recherche en Informatique et en Automatique - LMO - Laboratoire de Mathématiques d'Orsay - Université Paris-Saclay - CNRS - Centre National de la Recherche Scientifique) |
| Abstract: | We revisit the interest of classical statistical techniques for sales forecasting like exponential smoothing and extensions thereof (as Holt's linear trend method). We do so by considering ensemble forecasts, given by several instances of these classical techniques tuned with different (sets of) parameters, and by forming convex combinations of the elements of ensemble forecasts over time, in a robust and sequential manner. The machine-learning theory behind this is called "robust online aggregation", or "prediction with expert advice", or "prediction of individual sequences" (see Cesa-Bianchi and Lugosi, 2006). We apply this methodology to a hierarchical data set of sales provided by the e-commerce company Cdiscount and output forecasts at the levels of subsubfamilies, subfamilies and families of items sold, for various forecasting horizons (up to 6-week-ahead). The performance achieved is better than what would be obtained by optimally tuning the classical techniques on a train set and using their forecasts on the test set. The performance is also good from an intrinsic point of view (in terms of mean absolute percentage of error). While getting these better forecasts of sales at the levels of subsubfamilies, subfamilies and families is interesting per se, we also suggest to use them as additional features when forecasting demand at the item level. |
| Keywords: | prediction with expert advice,ensemble forecasts,exponential smoothing,Holt's linear trend method,e-commerce data |
| Date: | 2020–06–05 |
| URL: | http://d.repec.org/n?u=RePEc:hal:wpaper:hal-02794320&r=all |
| By: | Pietro Rossi; Flavio Cocco; Giacomo Bormetti |
| Abstract: | Building the future profit and loss (P&L) distribution of a portfolio holding, among other assets, highly non-linear and path-dependent derivatives is a challenging task. We provide a simple machinery where more and more assets could be accounted for in a simple and semi-automatic fashion. We resort to a variation of the Least Square Monte Carlo algorithm where interpolation of the continuation value of the portfolio is done with a feed forward neural network. This approach has several appealing features. Neural networks are extremely flexible regressors. We do not need to worry about the fact that for multi assets payoff, the exercise surface could be non connected. Neither we have to search for smart regressors. The idea is to use, regardless of the complexity of the payoff, only the underlying processes. Neural networks with many outputs can interpolate every single assets in the portfolio generated by a single Monte Carlo simulation. This is an essential feature to account for the P&L distribution of the whole portfolio when the dependence structure between the different assets is very strong like the case where one has contingent claims written on the same underlying. |
| Date: | 2020–06 |
| URL: | http://d.repec.org/n?u=RePEc:arx:papers:2006.09955&r=all |