2024. Unsupervised time-event probabilistic classification using large panels of time serienter (with Javier Palarea Albaladejo and Manuel Ruiz Marin). Download the paper. Download the R codes that simulates breaks in means, cycles in means, and breaks in variances. Download the R codes that replicate the examples for dating the business cycle turning points in the US from the state-level coincident indexes and the of periods of high volatility in the US stock market from the constituent assests of the SP500. | Abstract. This
work presents a general framework for partitioning a time span into
meaningful, non-overlapping segments using time series datasets
characterized by a large cross-sectional dimension. These datasets
commonly exhibit complexities and challenges such as non-linearities,
structural breaks, asynchronicity, missing data or significant outliers
that hamper data analysis and modelling. Aiming at accurate time-event
classification and change or breakpoint detection in this setting, our
proposal integrates three distinct components into a unified approach:
symbolic analysis, compositional data analysis, and Markov-switching
time series modeling. A comprehensive Monte Carlo simulation study is
conducted to assess the performance of the method, demonstrating
exceptional robustness across diverse scenarios. Moreover, its use in
real-world applications is illustrated through two economic examples:
(i) identifying recurrent recession and expansion regimes in the US
economic cycle; and (ii) dating change points to high volatility
episodes in the US stock market.
|
2024. Intertemporal restrictions in structural vector autoregressions (with Yuliya Lovcha). Download the paper and the online technical appendix. Download the Matlab codes that replicate the results. | Abstract. We
propose an innovate identification scheme, which we call intertemporal
identification, that focuses on the relative importance of identified
shocks at different frequencies to establish a structural
interpretation of the responses in VAR models. The method is highly
intuitive and can be used to identify structural shocks either
independently or in conjunction with other approaches, such as sign
restrictions. This is particularly useful when the latter approach is
questionable or generate wide sets of admissible responses.
Theoretically and through simulation-based scenarios, we show that the
intertemporal identifications tend to significantly reduce the
identified sets of sign restrictions and derive the conditions under
which one of the restrictions becomes redundant. We illustrate the
usefulness of our approach with three empirical examples: (i)
identification of technology shocks; (ii) identification of
oil-specific demand shocks; and (iii) identification of monetary policy
shocks. |
2024. Generalized ROC function: application to business cycle classification (with Salvador Ramallo and Andres Romeu). Download the paper. | Abstract. The
three-dimension Generalized ROC (GROC) function generalizes the ROC
function by enlarging the two-dimension components of the ROC function,
False Positive Rates (FPR) and True Positive Rates (TPR), with the
value of the thresholds (a). To assess classification performance, we
propose the Area of the GROC (AGROC) function, which measures the
difference between the area under the projection of GROC on the TPRxa
plane and the area under the projection of GROC on the FPRxa plane.
Using simulations, we show the better performance of AGROC over the
standard AUROC to measure classifier performance when the classifiers
are probabilities of recession. In the empirical section, we show the
good performance of our approach to address which probabilities of
recession, computed from Markov-switching models to US data of GDP
growth rates, best capture the NBER-referenced state of the business
cycle. |
2022. Spillover effects in international business cycles (with Gabriel
Perez Quiros and Matias Pacce). Working paper n. 2034 at the Central
Bank of Spain series and n. 2484 at the ECB series Download the paper. | Abstract.
To analyze the international transmission of business cycle
fluctuations, we propose a new multilevel dynamic factor model with a
block structure that (i) does not restrict the factors to being
orthogonal and (ii) mixes data sampled at quarterly and monthly
frequencies. By means of Monte Carlo simulations, we show the high
performance of the model in computing inferences of the unobserved
factors, accounting for the spillover effects, and estimating the
model’s parameters. We apply our proposal to data from the G7 economies
by analyzing the responses of national factors to shocks in foreign
factors and by quantifying the changes in national GDP expectations in
response to unexpected positive changes in foreign GDPs. Although the
share of the world factor as a source of the international transmission
of fluctuations is still signicant, this is partially absorbed by the
spillover transmissions. In addition, we document a pro-cyclical
channel of international transmission of output growth expectations,
with the US and UK being the countries that generate the greatest
spillovers and Germany and Japan being the countries that generate the
smallest spillovers. Therefore, policymakers should closely monitor the
evolution of foreign business cycle expectations. |
|