Riding the Waves: State-Switching Models and Market Resilience

by

Author: Denis Avetisyan

A new approach to portfolio construction leverages dynamic state-switching to navigate economic turbulence and enhance investment returns.

From January 1980 to September 2020, a mean-variance investor-demonstrating a relative risk aversion of three-experienced varying cumulative wealth depending on asset allocation strategies-including predictive-regression models such as PLS ($EPLSE^{\text{PLS}}$), PCA ($EPCAE^{\text{PCA}}$), forecast combination ($EFCE^{\text{FC}}$), the historical average (HA), or buy-and-hold-and corresponding state-switching models, all recursively estimated with information available through the forecast period, with downturns marked by NBER-dated recessions and a notable demarcation at the close of 2019.
From January 1980 to September 2020, a mean-variance investor-demonstrating a relative risk aversion of three-experienced varying cumulative wealth depending on asset allocation strategies-including predictive-regression models such as PLS ($EPLSE^{\text{PLS}}$), PCA ($EPCAE^{\text{PCA}}$), forecast combination ($EFCE^{\text{FC}}$), the historical average (HA), or buy-and-hold-and corresponding state-switching models, all recursively estimated with information available through the forecast period, with downturns marked by NBER-dated recessions and a notable demarcation at the close of 2019.

This review demonstrates that incorporating an Aligned Economic Index into state-switching models significantly improves portfolio performance and delivers substantial economic gains compared to traditional methods.

Predicting stock market returns remains a persistent challenge, particularly across varying economic conditions. This is addressed in ‘Switching between states and the COVID-19 turbulence’ which introduces a state-switching model and a novel Aligned Economic Index to improve return predictability. The analysis demonstrates that this approach significantly enhances portfolio performance and delivers substantial economic gains for investors, even during periods of market turbulence like the COVID-19 pandemic. Could this dynamic forecasting framework offer a more robust solution for asset allocation in an increasingly complex financial landscape?

The Inherent Fragility of Economic Prediction

Conventional economic forecasting fundamentally depends on the assumption of relatively stable relationships between variables – for example, a consistent link between interest rates and investment, or consumer confidence and spending. However, this premise is increasingly challenged by the dynamic and often unpredictable nature of modern economies. Shifts in technology, globalization, geopolitical events, and even collective consumer behavior can rapidly alter these established connections, rendering historical data less reliable as a predictor of future outcomes. Consequently, models built on past regularities may struggle to accurately capture the complexities of evolving economic landscapes, leading to forecasting errors and diminished predictive power. The inherent difficulty lies not in a lack of data, but in the fact that the very rules governing economic interactions are subject to change, demanding constant adaptation and a critical reassessment of established forecasting techniques.

Economic forecasting frequently falters not because of flawed theory, but due to the shifting sands of its underlying parameters. Models built on historical data assume a degree of consistency in relationships – that consumer behavior, market responsiveness, and external factors will remain relatively stable. However, the reality is often one of constant flux, where these parameters are rarely fixed and are subject to unpredictable shocks and evolving dynamics. This inherent instability introduces substantial model uncertainty, meaning that even the most sophisticated algorithms struggle to deliver reliable real-time predictions. The cumulative effect of these uncertainties can quickly render forecasts inaccurate, highlighting the limitations of relying solely on past data to anticipate future economic trends and underscoring the need for adaptive and robust forecasting strategies.

Despite their promise, time-varying forecasting models often struggle to deliver reliable predictions due to inherent statistical challenges. These models, designed to adapt to shifting economic landscapes, frequently fall prey to overfitting, where they learn the noise within the historical data rather than the underlying patterns. This results in excellent performance on past data but poor generalization to future, unseen data. Further compounding the issue is parameter instability; the estimated relationships within the model change rapidly over time, requiring constant recalibration. The frequent adjustments, intended to capture evolving dynamics, instead introduce further uncertainty and can amplify forecasting errors, ultimately diminishing the model’s predictive power and highlighting the difficulty of accurately forecasting in a constantly changing world.

From January 1980 to September 2020, dynamically adjusted equity allocations-optimized with varying forecasting methods (PLS, PCA, and forecast combination) and recursively updated data-responded to economic conditions as indicated by NBER recession dates, demonstrating a shift in strategy by the end of 2019.
From January 1980 to September 2020, dynamically adjusted equity allocations-optimized with varying forecasting methods (PLS, PCA, and forecast combination) and recursively updated data-responded to economic conditions as indicated by NBER recession dates, demonstrating a shift in strategy by the end of 2019.

Constructing a Resilient Foundation: The Aligned Economic Index

Traditional econometric forecasting often relies heavily on unrestricted estimation, potentially leading to unstable or unrealistic predictions. Economically motivated restrictions, however, introduce a priori knowledge and theoretical constraints into the modeling process. These constraints, derived from established economic principles – such as the law of demand or the quantity theory of money – reduce the parameter space and improve forecast accuracy by preventing the estimation of economically implausible values. This approach is particularly valuable when dealing with limited data or high-dimensional problems, where unrestricted models are prone to overfitting. By explicitly incorporating economic theory, these restrictions enhance the robustness and interpretability of forecasts, yielding more reliable predictions than purely data-driven methods.

The Aligned Economic Index (AEI) utilizes Partial Least Squares (PLS) regression to integrate fundamental economic predictors – variables with established theoretical links to economic activity – with premium information sources, such as high-frequency data or sentiment indicators. PLS is employed as it effectively handles multicollinearity inherent in many economic datasets and allows for dimensionality reduction, focusing on the most relevant predictive components. This technique differs from standard Ordinary Least Squares (OLS) by prioritizing the explanation of covariance in the predictors rather than solely minimizing residual variance, which is particularly beneficial when dealing with a large number of predictors and potential noise in premium data. The resulting AEI is a composite indicator derived from a weighted combination of these inputs, optimized to maximize predictive accuracy and stability.

The Aligned Economic Index utilizes a dual regression approach to differentiate between stable and changing economic relationships. One-State Regression is employed to model persistent effects, identifying predictors with consistently significant relationships to the target variable over the entire observed period. Complementing this, State-Switching Regression is implemented to detect dynamic effects, allowing the model to identify shifts in predictor significance and coefficient values that occur during specific sub-periods. This combination enables the Index to account for both long-term, fundamental drivers and shorter-term, regime-dependent influences on economic activity, improving its responsiveness to evolving conditions and potentially enhancing forecasting accuracy.

From Prediction to Performance: Asset Allocation Strategies

The Aligned Economic Index (AEI) functions as a predictive variable within a Mean-Variance Portfolio optimization process, determining asset weights based on forecasted economic conditions. This framework systematically adjusts portfolio allocation between the S&P 500 Index and US Treasury Bills to maximize the Sharpe Ratio, a measure of risk-adjusted return. Specifically, the AEI’s state – indicating expansion or recession – informs the optimization algorithm, shifting asset weights to favor equities during expansionary periods and fixed income during contractions. The resulting allocation aims to achieve a portfolio composition that offers the highest expected return for a given level of risk, as defined by portfolio volatility, and is dynamically rebalanced based on the AEI’s signal.

The asset allocation strategy centers on the S&P 500 Index and US Treasury Bills as primary asset classes. This selection allows for a diversified portfolio with exposure to both equity market gains and relatively stable, low-risk returns from US Treasury Bills. The core objective of this approach is to maximize the Sharpe Ratio, a metric calculated as the portfolio return above the risk-free rate divided by its standard deviation. By dynamically adjusting allocations between these two assets, the model aims to achieve the highest possible excess return per unit of risk, thereby optimizing portfolio efficiency and risk-adjusted performance. The selection of these assets simplifies model implementation while providing a representative benchmark for broader market exposure.

Evaluating the efficacy of dynamic investment strategies, including seemingly passive approaches like Buy-and-Hold, necessitates a comprehensive accounting of transaction costs. These costs, encompassing brokerage fees, bid-ask spreads, and potential market impact, directly reduce overall returns and can significantly distort performance metrics. Ignoring these expenses provides an overly optimistic assessment of strategy effectiveness; a strategy appearing profitable on a gross basis may become unattractive when net returns are considered. Accurate performance attribution requires the deduction of all associated transaction costs to provide a realistic evaluation of a strategy’s value proposition and to enable meaningful comparisons between different investment approaches.

Analysis indicates that implementation of a state-switching model, informed by the Aligned Economic Index, yielded annualized gains of up to 5.54% following the deduction of transaction costs. This performance metric reflects the net return achieved after accounting for all costs associated with portfolio rebalancing and trade execution. The methodology employed a dynamic asset allocation strategy, adjusting portfolio weights based on signals derived from the Aligned Economic Index, and the reported gain represents the overall profitability of this approach under the tested conditions.

A certainty-equivalent return (CER) measures the guaranteed return an investor would accept instead of taking on the risk of an investment. The study findings indicate that utilizing a state-switching model guided by the Aligned Economic Index generated a CER that was 183% higher than that achieved by a traditional buy-and-hold strategy. This substantial increase demonstrates the potential for improved risk-adjusted returns through dynamic asset allocation based on macroeconomic indicators, effectively translating predicted economic shifts into quantifiable investment gains exceeding those of a static approach.

The state-switching model, driven by the Aligned Economic Index, demonstrated a Sharpe Ratio of 0.25. This metric, calculated as the average return exceeding the risk-free rate per unit of volatility, signifies risk-adjusted return. This result is nearly double the 0.13 Sharpe Ratio achieved by the benchmark buy-and-hold strategy, indicating the state-switching model delivers a substantially improved return for each unit of risk taken. A higher Sharpe Ratio is generally preferred by investors as it suggests better compensation for the level of risk assumed.

The EPCAE (Economic and Policy Condition Assessment) model exhibits heightened performance during recessionary periods, delivering certainty-equivalent returns (CER) of up to 5.82%. This indicates a significant advantage in capital preservation and growth relative to benchmark strategies when economic conditions deteriorate. The observed CER gains during recessions suggest the model’s state-switching mechanism effectively identifies and capitalizes on shifts in economic regimes, allowing for a more responsive and potentially profitable investment strategy during downturns. This performance is a key driver of the model’s overall annualized gains and its superior Sharpe Ratio compared to a buy-and-hold approach.

The pursuit of optimized asset allocation, as detailed within this study, reveals a fundamental truth about complex systems. Every attempt to predict market behavior, to establish a stable baseline for return predictability, is inherently a negotiation with the ever-shifting present. This mirrors the inevitable decay all systems experience. As Albert Einstein observed, “The measure of intelligence is the ability to change.” The Aligned Economic Index, through its state-switching model, doesn’t attempt to prevent turbulence, but rather to adapt to it, acknowledging that change is the only constant. This adaptive approach, allowing portfolios to gracefully navigate different economic regimes, embodies a profound understanding of time’s influence on even the most carefully constructed financial strategies.

The Inevitable Shift

The demonstrated gains from incorporating a state-switching model, particularly one anchored by an Aligned Economic Index, are not anomalies. They represent the predictable outcome of acknowledging system dynamism. Traditional portfolio construction treats time as a linear progression, a metric against which returns are measured. This work subtly suggests the opposite: time is the medium within which states shift, and predictive power arises not from forecasting in time, but from recognizing where within the system the current state resides. The gains are merely the reduction of error as the model better maps this state space.

However, elegance does not equate to completion. The reliance on a specific economic index, while demonstrably effective, introduces a point of potential fragility. The system will inevitably encounter conditions where that index loses its alignment-a new variable emerges, or the relationships between existing variables decay. The next iteration of this work must address the question of index robustness-how to construct an index that ages gracefully, adapting to the subtle shifts in the underlying economic landscape without sacrificing predictive accuracy.

Ultimately, the pursuit of “optimal” allocation is a Sisyphean task. Systems degrade. What this research offers is not a solution, but a refined method for managing that degradation-a means of reducing the rate of error, and extending the period of functional stability. The true metric of success will not be absolute return, but the longevity of the model’s effectiveness in a world defined by perpetual change.

Original article: https://arxiv.org/pdf/2512.20477.pdf

Contact the author: https://www.linkedin.com/in/avetisyan/

See also:

2025-12-24 19:01