Tracking the economy at high frequency
summary
The gist
This paper develops a high-frequency economic indicator using a Bayesian Dynamic Factor Model estimated with mixed-frequency data to provide real-time economic monitoring in Ecuador…
In short
Researchers developed a High-Frequency Economic Index (HFEI) for Ecuador using a Bayesian Dynamic Factor Model to monitor the economy in real-time. This index successfully tracks business cycles and major shocks like COVID-19 and oil downturns, offering more timely insights than traditional quarterly GDP data.
Key concepts
- Bayesian Dynamic Factor Model (BDFM)
- A statistical model used to analyze complex economic data with different frequencies. It assumes that most variables move together due to a single underlying 'latent' factor, and it estimates the system's structure while accounting for changing volatility over time.
- Mixed-Frequency Data
- Data collected at different time intervals, such as weekly, monthly, and quarterly. The model is specifically designed to handle this mixed data type by creating a consistent 'pseudo-week' structure to harmonize observations for analysis.
- Pseudo-Weekly Index (HFEI)
- The final economic index created by the model. It is scaled to match the growth rate of real GDP and provides a high-resolution view of Ecuador's economic activity, capturing short-term fluctuations missed by slower official measures.
Terminology used across episodes
This episode discusses
The paper
Tracking the economy at high frequency · Read on arXiv
Universidad Ecotec
This paper develops a high-frequency economic indicator using a Bayesian Dynamic Factor Model estimated with mixed-frequency data. The model incorporates weekly, monthly, and quarterly official indicators, and allows for dynamic heterogeneity and stochastic volatility. To ensure temporal consistency and avoid irregular aggregation artifacts, we introduce a pseudo-week structure that harmonizes the timing of observations. Our framework integrates dispersed and asynchronous official statistics into a unified High-Frequency Economic Index (HFEI), enabling real-time economic monitoring even in environments characterized by severe data limitations. We apply this framework to construct a high-frequency indicator for Ecuador, a country where official data are sparse and highly asynchronous, and compute pseudo-weekly recession probabilities using a time-varying mean regime-switching model fitted to the resulting index.
Transcript
Introduction to the show: ident: Economics. The day in economics: theory, mechanisms and the evidence beneath them.
Ruth: We're looking at the paper: "Tracking the economy at high frequency".
Warren: Here's the gist of the paper. This paper develops a high-frequency economic indicator using a Bayesian Dynamic Factor Model estimated with mixed-frequency data to provide real-time economic monitoring in Ecuador, which is characterized by sparse and asynchronous official data The proposed index accurately tracks major phases of Ecuador’s business cycle, including global shocks like the 2008-09 financial crisis and the COVID-19 pandemic, as well as domestic episodes such as the 2015-16 oil-related downturn Data Construction and Harmonization The researchers construct a novel dataset composed entirely of official sources, integrating variables at weekly, monthly, and quarterly frequencies to create a high-frequency indicator for Ecuador This dataset includes seventeen series ranging from 2004 to the fourth quarter of 2024 across quarterly, monthly, and weekly indicators A key step involves introducing a pseudo-week structure that harmonizes the timing of observations to ensure temporal consistency and avoid irregular aggregation artifacts The definition of a pseudo-week is structured as dividing each month into four equal periods, ensuring intra-month consistency while allowing for weekly aggregation Model Specification and Estimation The core methodology employs a Bayesian dynamic factor model (BDFM) cast in state-space form to accommodate mixed-frequency data, dynamic heterogeneity, and time-varying volatility The model assumes a single latent weekly cyclical factor, ft, that captures most of the comovement between all variables in the system The estimation is performed using a Gibbs sampler within a Bayesian framework to jointly infer latent variables, time-varying volatilities, and missing observations via the Kalman filter and simulation smoother The model evaluates eight specifications varying along two dimensions: the degree of dynamic heterogeneity and the structure of stochastic volatility Index Interpretation and Performance The resulting High-Frequency Economic Index (HFEI), derived from the latent factor time series, is interpreted as a pseudo-weekly economic index scaled to match the mean and standard deviation of the year-over-year growth rate of real GDP The HFEI accurately tracks major phases of Ecuador’s business cycle, including global shocks like the 2008-09 financial crisis and the COVID-19 pandemic The index's real-time performance during the 2024 slowdown underscores its practical relevance for short-term monitoring Structural Complexity Trade-offs Model comparison using the Deviance Information Criterion (DIC) reveals that models with stochastic volatility in the idiosyncratic errors tend to produce smoother and more homogeneous cyclical movements The model with stochastic volatility in the factor and heterogeneous dynamics strikes a useful balance, exhibiting smoother transitions and more clearly defined turning points compared to models without dynamic heterogeneity While structural complexity is rewarded in terms of statistical fit, capturing economically meaningful dynamics requires careful consideration of the tradeoff between model flexibility and interpretability The authors suggest that incorporating stochastic volatility in the factor yields more interpretable and economically plausible estimates, even when not strictly favored by model selection criteria Recession Probability Estimation The latent factor time series, ft, can be decomposed into a time-varying mean µt and a noise term ǫt to serve as input to a second-stage Markov-switching model for estimating recession probabilities This regime-switching model distinguishes between heterogeneous recessions and expansions, allowing the depth and duration of each regime to vary over time The estimated recession probabilities provide a consistent and interpretable signal of the likelihood that the economy was in a recession at any given time The framework introduced here can be readily extended to other emerging markets as access to high-frequency administrative and survey data expands Comparison with Existing Measures The High-Frequency Economic Index, in contrast to official quarterly indices, leverages mixed-frequency data and aims to capture more nuanced and timely signals of economic dynamics beyond those embedded solely in GDP The HFEI's behavior toward the end of the sample documents a gradual slowdown over 2024 that is less abrupt than what is ultimately shown in quarterly GDP data The index captures nuances, such as the unusual resilience of deposits during the pandemic, which traditional indicators might obscure The HFEI provides substantially greater temporal resolution than GDP alone, anticipating and mirroring general direction while identifying short-lived fluctuations The HFEI and the associated recession probabilities document four well-defined episodes of economic contraction, which contrasts with the six contraction periods identified by GDP alone The authors emphasize that their indicators are less prone to such false signals thanks to their multivariate foundation and the use of a regime-switching model with time-varying mean Data Details and Methodology The construction of the dataset involved careful handling of missing observations, such as interpolating inadequate employment series using information from social security records for consistency Weekly series for private sector credit and deposits were reconstructed through automated extraction from the Boletín Monetario Semanal (BMS) publications Trade variables were constructed from transaction-level customs data, with restrictions placed on export records prior to 2012 to address distortions The model's state vector is defined by the combination of factor components and idiosyncratic errors across quarterly, monthly, and weekly observations The paper develops a new High-Frequency Economic Index (HFEI) using a Bayesian dynamic factor model estimated on mixed-frequency data to provide real-time economic monitoring in Ecuador The proposed index accurately tracks major phases of Ecuador’s business cycle, including global shocks like the 2008-09 financial crisis and the COVID-19 pandemic, as well as domestic episodes such as the 2015-16 oil-related downturn The framework introduced here can be readily extended to other emerging markets as access to high-frequency administrative and survey data expands References Almuzara, Martin, Katie Baker, Hannah O’Keeffe, and Argia Sbordone (2023): “The New York Fed Staff Nowcast 2.0,” New York FED Staff Technical Paper Alvarez, Rocio, Maximo Camacho, and Gabriel Perez-Quiros (2012): “Finite sample performance of small versus large scale dynamic factor models,” Antolin-Diaz, Juan, Thomas Drechsel, and Ivan Petrella (2017): “Tracking the slowdown in long-run GDP growth,” Review of Economics and Statistics, 99 (2), 343–356 Antolín-Díaz, Juan, Thomas Drechsel, and Ivan Petrella (2024): “Advances in nowcasting economic activity: The role of heterogeneous dynamics and fat tails,” Journal of Econometrics, 238 (2), 105634 Aruoba, S Borağan, Francis X Diebold, and Chiara Scotti (2009): “Real-time measurement of business conditions,” Journal of Business & Economic Statistics, 27 (4), 417–427 Bańbura, Marta, Domenico Giannone, Michele Modugno, and Lucrezia Reichlin (2013): “Now-casting and the real-time data flow,” in Handbook of economic forecasting, Elsevier, vol.
Ruth: 2, 195–237 Banco Central del Ecuador (2021a): “Indicadores líderes compuestos del ciclo económico para una evaluación temprana de la actividad económica en Ecuador,” Nota Técnica 87, Banco Central del Ecuador — (2021b): “Revelando el ciclo económico en Ecuador a través de un filtro de variables observables,” Nota Técnica 86, Banco Central del Ecuador Baumeister, Christiane, Danilo Leiva-León, and Eric Sims (2024): “Tracking weekly state-level economic conditions,” Review of Economics and Statistics, 106 (2), 483–504 Boivin, Jean and Serena Ng (2006): “Are more data always better for factor analysis?” Journal of Econometrics, 132 (1), 169–194 Bragoli, Daniela (2017): “Now-casting the Japanese economy,” International Journal of Forecasting, 33 (2), 390–402 Camacho, Maximo and Gabriel Perez-Quiros (2010): “Introducing the euro-sting: Shortterm indicator of euro area growth,” Journal of Applied Econometrics, 25 (4), 663–694 Carriero, Andrea, Todd E Clark, Massimiliano Marcellino, and Elmar Mertens (2024): “Addressing COVID-19 outliers in BVARs with stochastic volatility,” Review of Economics and Statistics, 106 (5), 1403–1417 Celeux, Gilles, Florence Forbes, Christian P Robert, and D Mike Titterington (2006): “Deviance information criteria for missing data models,”.
Warren: Journal of the royal statistical society: Series b (statistical methodology), 64 (4), 583–639 Chan, Joshua CC and Angelia L Grant (2016a): “Fast computation of the deviance information criterion for latent variable models,” Computational Statistics & Data Analysis, 100, 847–859 — (2016b): “On the observed-data deviance information criterion for volatility modeling,” Journal of Financial Econometrics, 14 (4), 772–802 Chan, Joshua CC, Aubrey Poon, and Dan Zhu (2023): “High-dimensional conditionally Gaussian state space models with missing data,” Journal of Econometrics, 236 (1), 105468 Chauvet, Marcelle and James D Hamilton (2006): “Dating business cycle turning points,” Contributions to Economic Analysis, 276, 1–54 D’Agostino, Antonello, Domenico Giannone, Michele Lenza, and Michele Modugno (2016): “Nowcasting business cycles: A Bayesian approach to dynamic heterogeneous factor models,” in Dynamic Factor Models, Emerald Group Publishing Limited, 569–594 Durbin, James and Siem Jan Koopman (2002): “A simple and efficient simulation smoother for state space time series analysis,” Biometrika, 89 (3), 603–616 García-Albán, Freddy, Manuel González-Astudillo, and Cristhian Vera-Avellán (2021): “Good policy or good luck?
Ruth: Analyzing the effects of fiscal policy and oil revenue shocks in Ecuador,” Energy Economics, 100, 105321 González-Astudillo, Manuel and Daniel Baquero (2019): “A nowcasting model for Ecuador: Implementing a time-varying mean output growth,” Economic Modelling, 82, 250–263 Hartwig, Benny (2024): “Bayesian VARs and prior calibration in times of COVID-19,” Studies in Nonlinear Dynamics & Econometrics, 28 (1), 1–24 Kim, Sangjoon, Neil Shephard, and Siddhartha Chib (1998): “Stochastic Volatility: Likelihood Inference and Comparison with ARCH Models,” The Review of Economic Studies, 65 (3), 361–393 Leiva-León, Danilo, Gabriel Perez-Quiros, and Eyno Rots (2024): “Real-time weakness of the global economy,” Journal of Applied Econometrics, 39 (5), 813–832 Lenza, Michele and Giorgio E Primiceri (2022): “How to estimate a vector autoregression after March 2020,” Journal of Applied Econometrics, 37 (4), 688–699 Lewis, Daniel J, Karel Mertens, James H Stock, and Mihir Trivedi (2022): “Measuring real activity using a weekly economic index,” Journal of Applied Econometrics, 37 (4), 667–687 Mariano, Roberto S and Yasutomo Murasawa (2003): “A new coincident index of business cycles based on monthly and quarterly series,” Journal of applied Econometrics, 18 (4), 427–443 Romero, Pedro, Rodrigo López, S., and Pablo Carrión (2025): “Índice Incertidumbre de Economía y Política de Ecuador: U-Index USFQ,” Available at ResearchGate Sargent, Thomas J, Christopher A Sims, et al.
Paper summary: Ruth: So, we’re looking at this paper from arXiv called "Tracking the economy at high frequency." It tackles the problem of getting timely economic data for Ecuador, a country where official statistics are really sparse and often come in big chunks like quarterly or monthly reports. The authors propose using a Bayesian Dynamic Factor Model estimated on mixed-frequency data to build this indicator.
Warren: So, what’s the main claim here? They’re trying to create a high-frequency economic index, which they call the High-Frequency Economic Index, HFEI. It’s designed for real-time monitoring of Ecuador's business cycle.
Iris: The thesis seems to be that by combining weekly, monthly, and quarterly official indicators—seventeen series from two thousand four through the fourth quarter of two thousand twenty-four—and using a Bayesian Dynamic Factor Model, you can create a unified index that works even when the data is messy and asynchronous <ref:2507.07450#pg1,weekly, monthly, and quarterly official indicators>.
Ruth: Exactly. They introduce something called a pseudo-week structure to harmonize all these different timeframes so they don't get messed up by the way data is collected. It’s about bridging that gap in official reporting, which is what makes this work important for monitoring things like global shocks or domestic events in Ecuador.
Warren: And they claim this index accurately tracks major phases of the business cycle, including big ones like the two thousand eight to two thousand nine financial crisis and the COVID-nineteen pandemic, as well as local episodes like that oil-related downturn in two thousand fifteen to two thousand sixteen <ref:2507.07450#pg3,the COVID-19 pandemic, as>. That’s a pretty strong claim for a country with limited data sources.
Iris: The results show that this HFEI provides substantially greater temporal resolution than just using official quarterly GDP data alone. For instance, the paper notes that the HFEI documents four well-defined episodes of economic contraction, which contrasts with six contraction periods identified by GDP figures alone.
Ruth: That difference in counting contraction periods is significant for understanding how quickly things are actually changing in Ecuador compared to what the standard reports show. But what about how they build this model? That’s where the technical details come in.
Warren: The methodology relies on a Bayesian dynamic factor model cast in state-space form, which is designed to handle that mixed frequency data, along with things like dynamic heterogeneity and time-varying volatility. They assume there’s one latent weekly cyclical factor capturing most of the co-movement between all these variables.
Paper summary: Iris: They use a Gibbs sampler within a Bayesian framework to jointly infer those latent variables, the time-varying volatilities, and even missing observations using techniques like the Kalman filter and simulation smoother. The model evaluates eight different specifications varying in how much dynamic heterogeneity is included and how stochastic volatility is structured in the error terms.
Ruth: That’s a lot of complexity being managed there. It sounds like they’re trying to balance fitting the data well with keeping the model interpretable, which is always a tough spot in these kinds of economic models. I wonder what happens when those underlying assumptions about dynamic heterogeneity change in real time, given how messy the data is supposed to be.
Warren: The paper does touch on that trade-off by comparing different specifications using the Deviance Information Criterion, or DIC. They found that models with stochastic volatility in the idiosyncratic errors tend to produce smoother and more homogeneous cyclical movements.
Iris: But they suggest that a model incorporating stochastic volatility in both the factor and heterogeneous dynamics strikes a useful balance, showing smoother transitions and more clearly defined turning points than models without dynamic heterogeneity. The authors hint that incorporating stochastic volatility in the factor actually yields estimates that are more interpretable and economically plausible even if those specific criteria don't strictly favor it.
Ruth: So, when you look at the results, this HFEI is scaled to match the mean and standard deviation of the year-over-year growth rate of real GDP. That gives us a concrete measure we can compare against official figures.
Warren: And then they take that latent factor time series and decompose it into a time-varying mean, denoted by mu t, and some noise term, epsilon t. This noise term is then fed into a second-stage Markov-switching model to estimate recession probabilities.
Iris: That’s the mechanism for the recession probability estimation. By using this regime-switching model, they can distinguish between different types of recessions—heterogeneous ones—and show how their depth and duration vary over time according to different regimes.
Ruth: For someone just listening, what does that mean practically? It means you get a consistent signal about the likelihood that the economy was in a recession at any specific point, based on this high-frequency index. It’s not just one big picture; it’s a probabilistic view over time.
Warren: And to put it in perspective, they compare their High-Frequency Economic Index with business cycle estimates published by the Central Bank of Ecuador based on real GDP and found a simple correlation coefficient of approximately zero. So, this new approach isn't just mirroring what the old indicators show; it’s capturing nuances that those traditional measures miss entirely.
Paper summary: Iris: It’s interesting how they handle data reconstruction too. They had to recover adequate employment series from administrative data by using linear interpolation guided by records from the Ecuadorian Social Security Institute for anything before September two thousand twenty and they reconstructed weekly private sector credit and deposits through automated extraction from the Boletín Monetario Semanal publications <ref:2507.07450#pg1>.
Ruth: That’s a lot of work just to get the raw ingredients ready for the model. It shows how much effort is required to make this high-frequency indicator work when you start with such disparate sources like administrative records and customs data.
Warren: And they also had to be careful about trade variables, restricting export records prior to two thousand twelve because of some distortions in that data, making sure the transaction-level customs data was used correctly <ref:2507.07450#pg1>. It’s a detailed setup for the measurement record, which is important for any quantitative trader looking at this.
Iris: The paper does acknowledge its limitations by noting that structural complexity can be rewarded statistically, but capturing economically meaningful dynamics requires careful consideration of the trade-off between model flexibility and interpretability. They also mention that extending this framework to other emerging markets will depend on access to high-frequency administrative and survey data.
Ruth: So, the main limitation they flag is that while the HFEI captures many nuances, it still relies on official data sources which are inherently limited in terms of frequency and consistency across different economic sectors. It stops where those underlying administrative or survey systems don't provide enough detail for a deeper look into economic activity.
Warren: So, looking at the whole picture of this paper, "Tracking the economy at high frequency," it’s about using advanced statistical modeling to build a real-time economic monitor for a data-scarce environment like Ecuador. It moves beyond just looking at quarterly GDP and tries to capture short-term fluctuations and shocks more accurately.
Iris: The implication is that for policymakers or investors in places with limited high-frequency data, having tools that can process mixed frequencies and incorporate stochastic volatility could provide much more timely insights into economic health than relying solely on the standard indicators we’re used to.
Ruth: And the authors suggest that this framework has potential to be readily extended to other emerging markets once they have access to better high-frequency administrative or survey data, which is a realistic path forward. It’s about creating a toolkit that can be adapted where the data allows it.
Conclusion: Ruth: So, we’re wrapping up this look at "Tracking the economy at high frequency." Basically, these authors took messy data from Ecuador—weekly, monthly, quarterly stuff—and built this index that lets you see what’s happening right now on a pseudo-weekly basis.
Warren: Yeah. It's about taking those official reports and creating something much faster to track the actual business cycle movements in real time. It's not just looking at the big quarterly numbers anymore, it’s trying to capture the short-term stuff too.
Iris: The main thing is they used a Bayesian dynamic factor model, which is this fancy statistical tool, to link all those different data frequencies together into one coherent picture. They built this index that actually tracks major shocks and domestic slowdowns pretty accurately.
Ruth: And what does that mean for the world? It means for places like Ecuador, where data flow is slow or spotty, they have a way to get a much clearer, real-time view of their economic health without waiting months for the next official release.
Warren: For me as a trader, it suggests we could be reacting to shifts faster because the signal isn't lagged by months. It’s about seeing trends developing sooner rather than just reacting when they finally show up in the quarterly reports.
Iris: From an econometric standpoint, I think what's interesting is how they handled that complexity with stochastic volatility—meaning the model lets the volatility itself change over time, which makes the index more flexible in capturing sudden shifts. But we have to remember those assumptions about how dynamic heterogeneity works are still key identifiers of what this index is actually measuring.
Ruth: Exactly, it's not a magic box. The authors admit that while their model fits the data well, you still have to weigh whether adding all that extra complexity actually makes the economic interpretation better or just makes the statistics look fancier.
Warren: And they did show how this index compares to what Ecuador’s central bank already publishes based on GDP, and it shows a zero correlation there. So it's not just repeating old stories; it’s finding something different in the noise.
Iris: Right, the implication is that for emerging markets that are wrestling with data access, this framework gives them a blueprint for building their own tools to monitor activity at a higher frequency. We need to see if that blueprint can actually be built on the ground elsewhere.
More episodes
- Daily Summary for 2026-11-19
- 2608.21274-Recommendation Quality and the Concentration of Consumption: Experimental Evidence from Netflix
- 2408.02757-A nonparametric test for diurnal variation in spot correlation processes
- 2306.16393-High-Dimensional Canonical Correlation Analysis
- 2605.17299-Geometric Brownian motion with intermittent entries and exits
- 2607.04753-Fooling Yourself: how narratives shape beliefs
- 2408.14671-Inference on Treatment Effects with Repeatedly Mismeasured High-Dimensional Controls
- 2201.06898-Difference-in-Differences Estimators for Treatments Continuously Distributed at Every Period
- 2501.15307-Influence Functions: A Unified Differentiation-Based Framework
- The day in conversation for 2026-11-19