Sunday, February 28, 2021

Improved Measurement of Real GDP

Tincho Almuzara, Gabriele Fiorentini, and Enrique Sentana have a fascinating new paper, "Aggregate Output Measurements: A Common Trend Approach,"

Totally reasonable to explore the cointegration route, and quite striking how much it reduces the extraction error variance for long-run objects.  At the same time, it’s remarkable how little difference it makes for short-run objects when the signal-to-noise ratio is high.

That is, the extractions in the left and right panels of their Fig 5 appear almost identical.  I would like to see a plot of their divergence (surely close to 0, although it would be interesting to see if/when the divergence is large -- e.g., are there business cycle effects?), along with a confidence interval (surely it would include 0?).  

Monday, February 22, 2021

Bridging Sparse Models and Factor Models

Check out the new "Bridging Factor and Sparse Models," by Fan, Masini, and Medeiros.  So many interesting connections to think about. 

A key distinction is hard constraints vs. soft constraints.  Both sparsity and factor structure (reduced rank) are examples of hard constraints, and in that sense they are more similar than different. Consider an  NxN parameter matrix M with elements a, b, ... Sparsity is the hard constraint M (or some subset of M) = 0, e.g., a = b = 0.  But one can of course consider more general hard constraints.  An obvious class is of the form  f(M) = 0, e.g., a + b^2 = 3.  Another obvious class, in multivariate environments, is reduced-rank (rank(M) < N) -- that is, factor structure.    

Wednesday, February 17, 2021

11th ECB Forecasting Conference

 Always a great conference!

11th ECB Conference on Forecasting Techniques — Macroeconomic forecasting in abnormal times,

will take place as an online event at the ECB on 15 and 16 June 2021.

Joshua Chan (Purdue University) and Christopher Sims (Princeton University) have confirmed their participation as keynote speakers. Lucrezia Reichlin (London Business School) will moderate a panel discussion with Joshua Chan, Christopher Sims and Matt Taddy (Amazon) on “Current challenges in macroeconomic forecasting and the way ahead”.

We particularly encourage submissions on robust forecasting in the presence of nonlinearities and structural breaks, such as COVID-19; forecasting inflation, real activity, and exchange rates under changing monetary policy frameworks; robust forecasting in the presence of large risks, e.g. related to climate change; modelling and forecasting when facing tail events; machine learning and scalable forecasting methods; big and/or unstructured data and dimension reduction techniques. However, the scope of the conference is not limited to these topics and submissions from all areas of forecasting are welcome.

The deadline for submissions is 23 March 2021.

You can find the call for papers and more details at

In the context of the conference, we are organising a paper competition for PhD students who have a research interest in forecasting. Please advertise this competition in your networks. The call for papers for this competition is online at

Best regards,

Elena BobeicaGabriel Pérez-QuirósGerhard Rünstler and Georg Strasser

Monday, February 15, 2021

Amazing Warming Paper, 1° × 1° Global Resolution!

Check out this paper. It is simultaneously (1) a sufficient statistic for much of the earlier empirical global warming economics research literature, and (2) a massive new contribution.

1° × 1° spatial resolution!

The Economic Geography of Global Warming
Jose Luis Cruz Alvarez and Esteban Rossi-Hansberg NBER w.p. #28466

Sunday, February 14, 2021

A Dynamic Factor Model for COVID Infections

True daily COVID infections are famously unobserved, but they are related to many observed noisy indicators, like  reported infections, reported deaths, hospitalizations, etc., with various leads and lags, various observational frequencies, etc. Sounds just like the ADS setup for business cycle nowcasting, doesn't it?  It would be great to see someone implement an ADS-style dynamic factor model for real-time monitoring of COVID infections.

Monday, February 8, 2021

AI, Machine Learning, and Big Data in Finance and Economics

An exciting new webinar series:

The AI & Big Data in Finance Research Forum (ABFR) is an interdisciplinary community of scholars interested in the methods, applications, and socioeconomic implications of AI and big data. ABFR organizes monthly presentations and discussions of papers from the leading experts in AI and Big Data in finance and economics. The virtual talks are the last Thursday of each month from 12-1pm EST. 

The first talk is February 25, 12-1pm EST.
Presenter: Stefan Nagel (University of Chicago)
Discussant: Kent Daniel (Columbia University)

The next talks are:
3/25 Laura Veldkamp discussed by Ezra Oberfield
4/29 Wei Jiang discussed by Franceso D’Acunto
5/27 Bryan Kelly discussant TBA.

For webinar information and Zoom links see 

To stay up to date, join the mailing list at

ABFR is organized by Svetlana Bryzgalova, Will Cong, Maryam Farboodi and Markus Pelger.  The hosting institutions are the Cornell FinTech Initiative and the Stanford AFTLab. The Advisory Committee includes Kay Giesecke, Gerald Hoberg, Wei Jiang, Bryan Kelly, Stefan Nagel, Andrew Patton, and Laura Veldkamp.

Monday, February 1, 2021

Machine Learning for Realized Volatility Forecasting

Check out this interesting paper.  Lots to think about.

Kim Christensen & Mathias Siggaard & Bezirgen Veliyev, 2021. "A machine learning approach to volatility forecasting," CREATES Research Papers 2021-03, Department of Economics and Business Economics, Aarhus University.

The paper shows that various ML methods outperform HAR for daily forecasting of realized asset-return volatility.  Of course there's nothing particularly interesting about HAR.  What's interesting is long memory, the overwhelmingly dominant feature of asset return realized vol dynamics, and HAR is just an approximate way to capture the long memory while staying in a comfortable linear regression framework.  Anyway, the key unanswered questions raised by the paper is how are the ML methods approximating long memory, and why do they deliver better approximations to long memory than HAR?  It is well known that long-memory and (infrequent) regime-switching are closely linked.  Perhaps the ML methods are picking up infrequent regime switching?  Do they also deliver better approximations than exact long-memory models?