Quant GANs: deep generation of financial time series

Abstract

Modeling financial time series by stochastic processes is a challenging task and a central area of research in financial mathematics. In this paper, we break through this barrier and present Quant GANs, a data-driven model which is inspired by the recent success of generative adversarial networks (GANs). Quant GANs consist of a generator and discriminator function which utilize temporal convolutional networks (TCNs) and thereby achieve to capture longer-ranging dependencies such as the presence of volatility clusters. Furthermore, the generator function is explicitly constructed such that the induced stochastic process allows a transition to its risk-neutral distribution. Our numerical results highlight that distributional properties for small and large lags are in an excellent agreement and dependence properties such as volatility clusters, leverage effects, and serial autocorrelations can be generated by the generator function of Quant GANs, demonstrably in high fidelity.

Neural networks for option pricing and hedging: a literature review

“…Finally, generative ANNs have been suggested recently as a non-parametric simulation tool for stock prices; see, for example, Henry-Labordère [2019], Kondratyev and Schwarz [2019], and Wiese et al [2019b]. Such simulation engines could then be used for option pricing and hedging, a direction still to be explored systematically.…”

Generating Realistic Stock Market Order Streams

“…Xiao et al (2017; introduced GAN-based methods for generating point processes; they generate the time for transaction events in stock markets. Other work aim to generate transaction prices in a stock market (Da Silva and Shi 2019; Koshiyama, Firoozye, and Treleaven 2019;Zhang et al 2019;Wiese et al 2019). Our problem is richer and harder as we aim to generate the actual limit orders including time, order type, price, and quantity information.…”

Deep Hedging: Learning to Simulate Equity Option Markets

“…Since DLVs are persistent we adopted ACF score proposed in [23]. It is defined by taking the Euclidean norm of the difference of the historical and the mean generated autocorrelation function…”

“…For comparison we apply the vector autoregressive model VAR(p) [14] and GAN-trained TCNs [23] to the same data. VAR(p) is a standard model for multivariate time series and assumes that X t+1 is an affine function of the past p observations and some Gaussian noise Z t+1 ∼ N (0, Σ):…”