Abstract:We address an important yet challenging problem - modeling high-dimensional dependencies across multivariates such as financial indicators in heterogeneous markets. In reality, a market couples and influences others over time, and the financial variables of a market are also coupled. We make the first attempt to integrate variational sequential neural learning with copula-based dependence modeling to characterize both temporal observable and latent variable-based dependence degrees and structures across non-normal multivariates. Our variational neural network WPVC-VLSTM models variational sequential dependence degrees and structures across multivariate time series by variational long short-term memory networks and regular vine copula. The regular vine copula models nonnormal and long-range distributional couplings across multiple dynamic variables. WPVC-VLSTM is verified in terms of both technical significance and portfolio forecasting performance. It outperforms benchmarks including linear models, stochastic volatility models, deep neural networks, and variational recurrent networks in cross-market portfolio forecasting.
What problem does this paper attempt to address?
The main problem that this paper attempts to solve is High - dimensional Cross - market Multivariate Dependence Modeling. Specifically, the author focuses on how to effectively model the high - dimensional dependence relationships among financial indicators in heterogeneous markets. The variables in these markets not only influence each other over time but also have complex coupling relationships with each other. The paper proposes a new method, that is, combining variational sequential neural learning with Copula - based dependence modeling, to capture the time - observable and latent variable dependencies and structures under non - normal multivariate conditions.
### Main contributions of the paper
1. **Proposed a new model**: WPVC - VLSTM (Weighted Partial Regular Vine Copula - based Variational LSTM), which is the first attempt to jointly model high - dimensional dependence, long - range dependence structures, non - normal time - multivariate, and latent features and relationships of heterogeneous market variables.
2. **Parameterized regular vine Copula**: Use a rich variational family Copula to better estimate the joint density of the posterior distribution.
3. **Avoid posterior collapse**: Through parameterization techniques, prevent the posterior collapse problem in variational auto - encoders (VAE).
4. **Efficient framework**: Combine vine Copula VAE with LSTM to capture the diverse dependencies and volatilities between different variables without imposing a Gaussian assumption on the multivariate.
### Problems solved
- **High - dimensional dependence modeling**: Traditional methods are prone to the curse of dimensionality problem when dealing with high - dimensional data. WPVC - VLSTM effectively solves this problem by combining variational neural learning and probability - dependence - structure modeling.
- **Non - normal distribution**: Data in financial markets usually does not conform to a normal distribution, and WPVC - VLSTM can capture the complex dependence relationships of non - normal distributions.
- **Long - range dependence**: LSTM itself can capture long - range dependence, but after combining with vine Copula, the model can handle complex dependence structures more flexibly.
- **Cross - market coupling**: The model can effectively model cross - market coupling relationships under complex market conditions, which is of great significance for cross - market portfolio management and risk control.
### Experimental verification
The paper experimentally verifies the performance of WPVC - VLSTM in cross - market portfolio prediction. The experimental results show that this model outperforms benchmark models such as linear models, stochastic volatility models, deep neural networks, and variational recurrent networks in both technical indicators and business impacts.
### Summary
By proposing the WPVC - VLSTM model, this paper successfully solves multiple challenges in high - dimensional cross - market multivariate dependence modeling, providing a powerful tool for cross - market portfolio management and risk control.