Машинно самообучение при VaR като оценка за пазарния риск ‒ предимства и недостатъци
VaR With Machine Learning as Market Risk Indicator - Advantages and Disadvantages

Summary/Abstract: The research paper endeavors to elucidate the predominant benefits and hurdles associated with employing Deep Autoregressive (DeepAR), a cutting-edge machine learning (ML) algorithm, for time series analysis, particularly in the estimation of Value at Risk (VaR). The investigation further encompasses an evaluation using Generalized Autoregressive Conditional Heteroscedasticity (GARCH) models as a widely recognized and longstanding quantitative method prevalent within the financial domain. The derived valuation covers currency risk (with an open currency position in JPY, USD and GBP) and equity position risk (long position in Apple and Microsoft Corporation). The juxtaposition of outcomes derived from the two models substantiates the quantitative ascendancy of the machine learning methodology. Nonetheless, the study also underscores several pragmatic challenges. Consequently, the application of machine learning techniques is deemed appropriate for deriving VaR estimates, contingent upon a comprehensive evaluation of the incremental benefits vis-à-vis the concomitant challenges.

• Details
• Contents