In my research on the economies and financial markets of developing countries, I have repeatedly encountered market behavior that is difficult to reconcile with the assumptions of conventional financial theory.

Inflation and interest rates can reach exceptionally high levels, exchange rates may undergo abrupt and substantial fluctuations, and asset prices often respond to several interacting factors at once. In commodity-dependent economies, these dynamics are further affected by recurring cycles in the international prices and trade volumes of primary commodities. Monetary policy changes, capital flows, and political or institutional factors can also alter the relationships between financial variables within relatively short periods.

A further difficulty is the quality of the available data. Missing observations, measurement errors, inconsistent statistical definitions, short sample periods, and changes in data collection methods are common. In some cases, it is difficult to determine whether an observed change reflects an actual change in market behavior or simply a problem with the underlying data.

I have explored a range of modeling approaches, including the Black–Scholes framework and its fractional extensions, stochastic volatility models, stochastic delay differential equations, and jump-diffusion processes. However, the introduction of more sophisticated stochastic structures has not necessarily resulted in better empirical performance. The assumptions underlying these models often do not adequately reflect the market conditions under investigation. In particular, parameter instability, non-stationarity, structural breaks, and unreliable observations make both estimation and model validation difficult.

As a result, much of the practical work has had to focus on data cleaning and validation, descriptive statistics, time-series analysis, regression methods, and decision-tree models. These approaches can provide a rough representation of market behavior, but they do not always offer a satisfactory theoretical explanation of the underlying economic mechanisms.

I have encountered similar difficulties when applying modern portfolio theory. Mean-variance optimization becomes questionable when asset returns exhibit unstable correlations, when exchange-rate and inflation shocks affect several assets simultaneously, and when liquidity constraints prevent investors from adjusting their positions freely. Under these conditions, the diversification benefits implied by historical covariance estimates may be misleading.

What concerns me most is not simply the failure of a particular model to produce accurate forecasts. It is the difficulty of establishing a reliable relationship between the statistical properties of the observed data, the underlying economic mechanisms, and the assumptions required for financial modeling.

I would therefore like to ask:

What theoretical and empirical approaches are currently being developed to address financial modeling under conditions of severe data limitations and structural instability, particularly in emerging and developing markets?

I am especially interested in the following questions:

Statistical inference under poor data quality: How can parameter estimation and statistical inference be made reliable when observations contain substantial measurement errors, missing values, inconsistent definitions, or limited historical coverage? To what extent can robust statistics, Bayesian methods, or state-space approaches address these problems?

Non-stationarity and changing market structures: How should financial models account for changes in the data-generating process itself? Are regime-switching models, time-varying parameter models, fractional stochastic processes, or other approaches capable of capturing these dynamics without introducing excessive estimation uncertainty?

Model risk and empirical validation: What validation framework is appropriate when the assumptions of stationarity and parameter stability cannot be taken for granted? In particular, how can we distinguish genuine predictive performance from spurious relationships, structural breaks, or overfitting?

Portfolio optimization under unstable dependence structures: What alternatives to conventional mean-variance optimization are most defensible when correlations are time-varying, tail dependence is significant, and market liquidity is limited? Can robust portfolio optimization, distributionally robust methods, or Expected Shortfall-based approaches provide more reliable results?

I am aware that these are related but distinct research problems. My main interest, however, is whether they can be addressed within a coherent modeling framework rather than through a collection of separate statistical adjustments.

I would particularly appreciate references to theoretical work, empirical studies, or practical methodologies that have been developed for these conditions. I am interested not only in improving predictive accuracy, but also in understanding how the resulting models can be interpreted economically and used in actual investment or policy decisions.

Quant

Message History

Delivery option(s)