Abstract
Dynamical systems comprising multiple components that can be partitioned into distinct blocks originate in many scientific areas. A pertinent example is the interactions between financial assets and selected macroeconomic indicators, which has been studied at aggregate level—e.g., a stock index and an employment index—extensively in the macroeconomics literature. A key shortcoming of this approach is that it ignores potential influences from other related components (e.g., Gross Domestic Product) that may exert influence on the system's dynamics and structure and thus produces incorrect results. To mitigate this issue, we consider a multi-block linear dynamical system with Granger-causal ordering between blocks, wherein the blocks' temporal dynamics are described by vector autoregressive processes and are influenced by blocks higher in the system hierarchy. We derive the maximum likelihood estimator for the posited model for Gaussian data in the high-dimensional setting based on appropriate regularization schemes for the parameters of the block components. To optimize the underlying non-convex likelihood function, we develop an iterative algorithm with convergence guarantees. We establish theoretical properties of the maximum likelihood estimates, leveraging the decomposability of the regularizers and a careful analysis of the iterates. Finally, we develop testing procedures for the null hypothesis of whether a block "Granger-causes" another block of variables. The performance of the model and the testing procedures are evaluated on synthetic data, and illustrated on a data set involving log-returns of the US S&P100 component stocks and key macroeconomic variables for the 2001–16 period.
Introduction
In fields like econometrics, genomics, and finance, researchers often deal with systems of many variables that naturally group together into hierarchical structures. Traditional Vector Autoregressive (VAR) models struggle in these "high-dimensional" settings because the number of parameters to estimate grows quadratically with the number of variables, often exceeding the available data points. Furthermore, many systems exhibit a "block-triangular" structure—for example, the stock market may influence macroeconomic variables like employment, but the reverse influence may occur at a different scale or not at all. Standard models often ignore this structure or rely on aggregate indices, which can mask the rich interactions between individual components. This paper provides a mathematically rigorous framework to estimate and test these multi-block relationships.
The Multi-Block VAR Framework
The proposed model utilizes a recursive structure where variables are organized into hierarchical blocks. In this framework, each block's evolution is determined by its own past and the past of blocks higher in the hierarchy.
A major insight of this recursive system is that the stability of the entire global system depends solely on the stability of the individual blocks. This allows the "cross-talk" between blocks to be complex without risking system-wide instability.
Estimation and Optimization
To handle high dimensionality, the authors use a penalized Maximum Likelihood Estimator (MLE) with two key regularizers:
- Lasso (L1) Penalty: Encourages sparsity within blocks to identify the most significant internal connections.
- Nuclear Norm Penalty: Allows for low-rank structures in the interactions between blocks, reflecting the idea that a block might react to an aggregate signal (like a latent portfolio) rather than every individual variable in another block.
Because the resulting likelihood function is non-convex, the authors developed a block-coordinate descent algorithm that alternates between estimating transition matrices and noise covariance matrices, backed by proven convergence guarantees.
Empirical Results
The model was validated on synthetic data and applied to a dataset of 71 S&P 100 stocks and 20 macroeconomic variables from 2001–2016.
- Granger-Causality: Testing strongly rejected the null hypothesis of no interaction, confirming that the stock market significantly "Granger-causes" the macroeconomy.
- The "Portfolio" Effect: The interaction between the stock market and the macroeconomy was found to be rank-1. This suggests the macroeconomy reacts to a single aggregate signal from the market rather than individual stock fluctuations.
- Crisis Connectivity: The model captured dramatic increases in connectivity during the 2008 financial crisis, identifying specific entities like AIG and Goldman Sachs as having outsized influence during the period.
- Forecasting: The multi-block model outperformed baseline models in one-step-ahead forecasting for macroeconomic variables by leveraging the lead-lag relationship from the stock market.
Conclusion
This framework provides a robust method for modeling high-dimensional dynamical systems with known hierarchical structures. While the current model assumes the hierarchy is known, it offers significant improvements in forecasting and structural analysis over traditional aggregate methods. Future work will explore methods to discover these hierarchies automatically and improve sensitivity for detecting weaker signals in smaller datasets.
Summary updated