Paper: arXiv 2404.01451

Authors: Katalin Varga, Tibor Szendrei

Abstract

Tracking the build-up of financial vulnerabilities is a key component of financial stability policy. Due to the complexity of the financial system, this task is daunting, and there have been several proposals on how to manage this goal. One way to do this is by the creation of indices that act as a signal for the policy maker. While factor modelling in finance and economics has a rich history, most of the applications tend to focus on stationary factors. Nevertheless, financial stress (and in particular tail events) can exhibit a high degree of inertia. This paper advocates moving away from the stationary paradigm and instead proposes non-stationary factor models as measures of financial stress. Key advantage of a non-stationary factor model is that while some popular measures of financial stress describe the variance-covariance structure of the financial stress indicators, the new index can capture the tails of the distribution. To showcase this, we use the obtained factors as variables in a growth-at-risk exercise. This paper offers an overview of how to construct non-stationary dynamic factors of financial stress using the UK financial market as an example.

Complexity vs Empirical Score

  • Math Complexity: 7.0/10
  • Empirical Rigor: 6.0/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper employs advanced statistical methodologies like dynamic Bayesian non-stationary factor models and limit theorems to handle spurious correlations, indicating high mathematical complexity. It is heavily empirical, applying these methods to real-world UK financial data, performing a growth-at-risk backtest, and benchmarking against existing indices like CLIFS and CISS.

Research Flowchart

  flowchart TD
  A["Research Goal:<br>Model Non-Stationary Financial<br>Vulnerability for the UK"] --> B["Data Inputs:<br>UK Financial Market<br>Stress Indicators"]
  B --> C["Methodology:<br>Non-Stationary Dynamic<br>Factor Model"]
  C --> D["Computational Process:<br>Estimate Factors<br>Capturing Tail Distributions"]
  D --> E["Outcome:<br>UK Financial Stress Index<br>Extremes identified"]
  D --> F["Application:<br>Growth-at-Risk Exercise"]
  F --> G["Key Finding:<br>Non-stationary factors provide<br>superior vulnerability signals<br>vs. stationary models"]
  E --> G

Evidence Card

What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card

Proposes

The paper proposes using non‑stationary factor models to build a UK Financial Stress Index that captures tails and inertia of financial stress.

Source sentence
This paper advocates moving away from the stationary paradigm and instead proposes non-stationary factor models as measures of financial stress.

Assumes 4

  • Financial stress exhibits high inertia and should be modelled with non‑stationary factors rather than stationary ones. model
    Source sentence
    The key advantage of non-stationary factor modelling for financial stability, is that financial stress is characterised by a high degree of inertia.
  • The time series used are non‑stationary (integrated) so that a generalized empirical covariance can be defined. data
    Source sentence
    In particular, the empirical mean and variance of non-stationary time series need not be constant
  • A limit theorem relating theoretical and empirical covariance exists for persistent n‑order integrated data. statistical
    Source sentence
    In the case of persistent, n-order integrated data, there exists the so-called generalised empirical covariance, which can be related to the theoretical covariance via a limit theorem
  • The optimal number of factors can be selected using the test described in Pe˜na and Poncela (2006). model
    Source sentence
    In particular, we show that it is optimal to pool all the chosen variables together and let the method select the optimal number of factors given the test described in Pe˜na and Poncela (2006).

Data

  • UK financial system variables (multiple stress indices) macro unknown
    UK financial market
    Source sentence
    We will then use the non-stationary factor modelling framework to construct the UK Financial Stress Index (UKFSI) for the United Kingdom using several variables pertaining to the UK financial system.
  • Country‑Level Index of Financial Stress (CLIFS) macro unknown
    UK
    Source sentence
    the availability of multiple stress indices constructed specifically for the UK economy, such as the Country‑Level Index of Financial Stress (CLIFS), and Sovereign Composite Indicator of Systemic Stress (SovCISS)
  • Sovereign Composite Indicator of Systemic Stress (SovCISS) macro unknown
    UK
    Source sentence
    ... Sovereign Composite Indicator of Systemic Stress (SovCISS)
  • Monthly macro variables (retail sales, inflation, industrial production, base rate, short‑ macro unknown
    UK
    Source sentence
    We use the monthly explanatory variables of Schorfheide and Song (2015) and Koop et al. (2023): retail sales, inflation, industrial production, base rate, short‑term interest rates, long‑term interest rates, and stock prices.
  • Monthly GDP series (mixed‑frequency estimates) macro unknown
    UK
    Source sentence
    Figure 1: The different measures of monthly GDP

Costs

Not applicable The paper reports no trading or execution result.

Finds 3

  • Factor‑based stress measures outperform other indices at short‑run forecast horizons, while SovCISS is best at longer horizons and CLIFS performs worst, especially at short horizons. performance
    Source sentence
    the overall message that factor models perform better at shorter forecast horizons remains intact, with the SovCISS only being the best model for half a year ahead forecast onwards.
  • Aggregating the factors using a simple arithmetic average reduces model performance compared to using the combined factors directly. performance
    Source sentence
    We also highlight that aggregating the factors yields worse performance for the model. We note however, that in this paper we used a simple arithmetic average to obtain aggregated factors
  • The proposed non‑stationary factor model shows lower predictive accuracy than SovCISS at longer forecast horizons. predictive-accuracy
    Source sentence
    our proposed method fairs worse than the SovCISS at longer forecast horizons

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 17 related papers). Coverage is partial, so this is neither replication nor refutation.

Code

No repository by the authors found in the paper.

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.

Cited References (48)

  1. Bayesian dynamic factor models and portfolio allocation (2000)
  2. The use of factor analysis in the statistical analysis of multiple time series (1963)
  3. Distribution of eigenvalues in multivariate statistical analysis (1983)
  4. Limiting distributions of least squares estimates of unstable autoregressive processes (1988)
  5. Time series analysis by state space methods (2012)
  6. Cointegration and common factors (1994)
  7. Principal components analysis of cointegrated time series (1997)
  8. The analysis of nonstationary time series using regression, correlation and cointegration (2012)
and 40 more
  1. Identifying a simplifying structure in time series (1987)
  2. Forecasting with nonstationary dynamic factor models (2004)
  3. Nonstationary dynamic factor analysis (2006)
  4. Introduction to econometrics. Boston: Addison-Wesley. (2003)
  5. Asymptotic properties of multivariate nonstationary processes with applications to autoregressions (1990)
  6. Nearly unstable AR models with coefficient matrices in Jordan normal form (1998)
  7. Why do we sometimes get nonsense-correlations between Time-Series?–a study in sampling and the nature of time-series (1926)
  8. FISS–A Factor-based Index of Systemic Stress in the Financial System (2020)
  9. Revisiting vulnerable growth in the Euro Area: Identifying the role of financial conditions in the distribution (2023)
  10. Co-integration and error correction: representation, estimation, and testing (1987)
  11. Macroeconomic data transformations matter (2021)
  12. Modelling non-stationary ‘big data’ (2021)
  13. Nonlinear factor models for network and panel data (2021)
  14. Sources and types of big data for macroeconomic forecasting (2020)
  15. Factor extraction using Kalman filter and smoothing: this is not just another survey (2021)
  16. The twin crises: the causes of banking and balance-of-payments problems (1999)
  17. CISS-a composite indicator of systemic stress in the financial system (2012)
  18. Systemic financial stress and macroeconomic amplifications in the United Kingdom (2022)
  19. Vulnerable growth (2019)
  20. Quantile autoregression (2006)
  21. Vulnerable growth in the euro area: Measuring the financial conditions (2020)
  22. Regression quantiles (1978)
  23. Comparing density forecasts using threshold-and quantile-weighted scoring rules (2011)
  24. High-frequency monitoring of growth at risk (2022)
  25. Mixed-frequency Growth-at-Risk with the MIDAS-QR method: Evidence from China (2023)
  26. Interquantile shrinkage in regression models (2013)
  27. Interquantile shrinkage and variable selection in quantile regression (2014)
  28. Nowcasting ‘true’monthly US GDP during the pandemic (2021)
  29. Reconciled estimates of monthly GDP in the United States (2023)
  30. Nowcasting in a pandemic using non-parametric mixed frequency VARs (2023)
  31. Real-time forecasting with a (standard) mixed-frequency VAR during a pandemic (2021)
  32. Real-time forecasting with a mixed-frequency VAR (2015)
  33. Dating systemic financial stress episodes in the EU countries (2017)
  34. Financial stress: What is it, how can it be measured, and why does it matter (2009)
  35. Collective risk management in a flight to quality episode (2008)
  36. Manias, panics, and crashes: A history of financial crises (2015)
  37. Information, liquidity, and the (ongoing) panic of 2007 (2009)
  38. Systemic risk: A survey (2000)
  39. Measuring financial stress in a developed country: An application to Canada (2006)
  40. Alternative tests for correct specification of conditional predictive densities (2019)

0 of 48 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.