I like thinking about causal estimation as a type of prediction (e.g., here). Here's a very nice slide deck from Peter Buhlmann at ETH Zurich detailing his group's recent and ongoing work in that tradition.

Some months ago I blogged on Google's new development of a dataset search tool.  Evidently it's coming along.  Check out the beta version here. Also, on dataset supply as opposed to demand, see here for how to maximize visibility of your datasets to the search engine.

[With thanks to the IIF's Oracle newsletter for alerting me.]

Last week's 13th annual Greater New York Area Econometrics Colloquium, generously hosted by Princeton, was a great success, with strong papers throughout. The program is below. I found two papers especially interesting. I already blogged on Spady and Stouli's “Simultaneous Mean-Variance Regression”. The other was "Nonparametric Sample Splitting", by Lee and Wang.

Richard Spady and Sami Stouli have an interesting new paper, “Dual Regression". They change the usual OLS loss function from quadratic to something related but different, as per their equation (2.2), and they get impressive properties for estimation under correct specification. They also have some results under misspecification.

I'd like to understand more regarding dual regression's properties for prediction under misspecification. Generally we're comfortable with quadratic loss, in which case OLS delivers the goods (the conditional mean or linear projection) in large samples under great generality (e.g., see here). The dual regression estimator, in contrast, has a different probability limit under misspecification -- it's not providing a KLIC-optimal approximation.

If the above sounds negative, note well that the issue raised may be an opportunity, not a pitfall! Certainly there is nothing sacred about quadratic loss, even if the conditional mean is usually a natural predictor. We sometimes move to absolute-error loss (conditional median predictor), check-function loss (conditional quantile predictor), or all sorts of other predictive loss functions depending on the situation. But movements away from conditional mean or median prediction generally require some justification and interpretation. Equivalently, movements away from quadratic or absolute predictive loss generally require some justification and interpretation. I look forward to seeing that for the loss function that drives dual regression.

The beautiful idea has been around for ages. Find the N closest H-histories to the current H-history (you choose/tune N and H), for each H-history see what followed, take an average, and use that as your forecast. Of course there are many variations and extensions. Interesting new work by Dendramis, Kapetanios, and Marcellino is in exactly that tradition, except that Dendramis et al.  don't show much awareness of the tradition, or attempt to stand on its shoulders, which I find odd. I find myself hungry for tighter connections, for example to my favorite old nearest-neighbor prediction piece, Sid Yakowitz's well-known "Nearest-Neighbor Methods for Time Series Analysis,” Journal of Time Series Analysis, 1987.

No, I have not gone into seclusion. Well actually I have, but not intentionally and certainly not for lack of interest in the blog. Just the usual crazy time of year, only worse this year for some reason. Anyway I'll be back very soon, with lots to say! But here's something important and timely, so it can't wait:

Journal of Financial Econometrics

Call for Papers

Special Issue in Honor of Peter Christoffersen

The Journal of Financial Econometrics is organizing a special issue in memory of Professor Peter
Christoffersen, our friend and colleague, who passed away in June 2018. Peter held the TMX Chair in Capital Markets and a Bank of Canada Fellowship and was a widely respected member of the Rotman School at the University of Toronto since 2010. Prior to 2010, Peter was a valued member of the Desautels Faculty of Management at McGill University. In addition to his transformative work in econometrics and volatility models, financial risk and financial innovation had been the focus of Peter’s work in recent years.

We invite paper submissions on topics related to Peter’s contributions to Finance and Econometrics. We are particularly interested in papers related to the following topics:

1)   The use of option-implied information for forecasting; Rare disasters and portfolio
management; Factor structures in derivatives and futures markets.

2)   Volatility, correlation, extreme events, systemic risk and Value-at-Risk modeling for
financial market risk management.

3)   The econometrics of digital assets; Big data and Machine Learning.

To submit a paper, authors should login to the Journal of Financial Econometrics online submission system and follow the submission instructions as per journal policy.  The due date for submissions is June 30, 2019.  It is important to specify in the cover letter that the paper is submitted to the special issue in honor of Peter Christoffersen, otherwise your paper will not be assigned to the guest editors.

Guest Editors

•    Francis X. Diebold, University of Pennsylvania

•    René Garcia, Université de Montréal and Toulouse School of Economics

•    Kris Jacobs, University of Houston

I just returned from a great BFI Conference at U Chicago, Developing and Using Business Expectations Data, organized by Nick Bloom and Steve Davis.

Wonderfully, density as opposed to point survey forecasts were featured throughout. There was the latest on central bank surveys (e.g., Binder et al.), but most informative (to me) was the emphasis on surveys that I'm less familiar with, typically soliciting density expectations from hundreds or thousands of C-suite types at major firms. Examples include Germany's important IFO survey (e.g., Bachman et al.), the U.S. Census Management and Organizational Practices Survey (e.g., Bloom et al.)., and fascinating work in progress at FRB Atlanta. 

The Census survey is especially interesting due to its innovative structuring of histogram bins. There are no fixed bins. Instead users give 5 bins of their own choice, and five corresponding probabilities (which add to 1). This solves the problem in fixed-bin surveys of  (lazy? behaviorally-biased?) respondents routinely and repeatedly assigning 0 probability to subsequently-realized events.

As the expansion ages, there's progressively more discussion of whether its advanced age makes it more likely to end. The answer is no. More formally, postwar U.S. expansion hazards are basically flat, in contrast to contraction hazards, which are sharply increasing. Of course the present expansion will eventually end, and it may even end soon, but its age it unrelated to its probability of ending.

All of this is very clear in Diebold, Rudebusch and Sichel (1992). See Figure 6.2 on p. 271. (Sorry for the poor photocopy quality.) The flat expansion hazard result has held up well (e.g., Rudebusch (2016)), and moreover it would only be strengthened by the current long expansion.

[I blogged on flat expansion hazards before, but the message bears repeating as the expansion continues to age.]

I am sad to report that Herman Stekler passed away last month. I didn't know until now. He was a very early and important and colorful -- indeed unique -- personage in the forecasting community, making especially noteworthy contributions to forecast evaluation.  
https://forecasters.org/herman-stekler_oracle-oct-2018/

Check out:

PUBLISHING AND PROMOTION IN ECONOMICS: THE TYRANNY OF THE TOP FIVE 
by
James J. Heckman and Sidharth Moktan 
NBER Working Paper 25093
http://www.nber.org/papers/w25093

Heckman et al. examine a range of data from a variety of perspectives, analyze them thoroughly, and pull no punches in describing their striking results.

It's a great paper. There's a lot I could add, maybe in a future post, but my blood pressure is already high enough for today. So I'll just leave you with a few choice quotes from the paper ["T5" means "top-5 economics journals" ]:

"The results ... support the hypothesis that the T5 influence operates through channels that are independent of article quality."

"Reliance on the T5 to screen talent incentivizes careerism over creativity."

"Economists at highly ranked departments with established reputations are increasingly not publishing in T5 or field journals and more often post papers online in influential working paper series, which are highly cited, but not counted as T5s."

"Many non-T5 articles are better cited than many articles in T5 journals. ...  Indeed, many of the most important papers published in the past 50 years have been too innovative to survive the T5 gauntlet."

"The [list of] most cited non-T5 papers reads like an honor roll of economic analysis."

"The T5 ignores publication of books. Becker’s Human Capital
(1964) has more than 4 times the number of citations of any paper listed on RePEc. The exclusion of books from citation warps incentives against broad and integrated research and towards writing bite-sized fragments of ideas."

Art Owen and Hal Varian have an eye-opening new paper, "Optimizing the Tie-Breaker Regression Discontinuity Design".

Randomized controlled trials (RCT's) are clearly the gold standard in terms of statistical efficiency for teasing out causal effects. Assume that you really can do an RCT. Why then would you ever want to do anything else?

Answer: There may be important considerations beyond statistical efficiency. Take the famous "scholarship example". (You want to know whether receipt of an academic scholarship causes enhanced academic performance among strong scholarship test performers.) In an RCT approach you're going to give lots of academic scholarships to lots of randomly-selected people, many of whom are not strong performers. That's wasteful. In a regression discontinuity design (RDD) approach ("give scholarships only to strong performers who score above X in the scholarship exam, and compare the performances of students who scored just above and below X"), you don't give any scholarships to weak performers. So it's not wasteful -- but the resulting inference is statistically inefficient. 

"Tie breakers" implement a middle ground: Definitely don't give scholarships to bottom performers, definitely do give scholarships to top performers, and randomize for a middle group. So you gain some efficiency relative to pure RDD (but you're a little wasteful), and you're less wasteful than a pure RCT (but you lose some efficiency).

Hence there's an trade-off, and your location on it depends on the size of the your middle group. Owen and Varian characterize the trade-off and show how to optimize the size of the middle group. Really nice, clean, and useful.

[Sorry but I'm running way behind. I saw Hal present this work a few months ago at a fine ECB meeting on predictive modeling.]

A few weeks ago the NBER released WP25000, What a great NBER service -- there have been 7.6 million downloads of NBER WP's in the last year alone. 

This milestone is of both current and historical interest. The history is especially interesting. As Jim Poterba notes in a recent communication:

This morning's "New this Week" email included the release of the 25000th NBER working paper, a study of the intergenerational transmission of human capital by David Card, Ciprian Domnisoru, and Lowell Taylor.  The NBER working paper series was launched in 1973, at the inspiration of Robert Michael, who sought a way for NBER-affiliated researchers to share their findings and obtain feedback prior to publication.  The first working paper was "Education, Information, and Efficiency" by Finis Welch.  The design for the working papers -- which many will recall appeared with yellow covers in the pre-digital age -- was created by H. Irving Forman, the NBER's long-serving chart-maker and graphic artist.

Initially there were only a few dozen working papers per year, but as the number of NBER-affiliated researchers grew, particularly after Martin Feldstein became NBER president in 1977, the NBER working paper series also expanded.  In recent years, there have been about 1150 papers per year.  Over the 45 year history of the working paper series, the Economic Fluctuations and Growth Program has accounted for nearly twenty percent (4916) of the papers, closely followed by Labor Studies (4891) and Public Economics (4877).

Very cool NYT graphics summarizing U.S. Facebook network connectedness.  Check it out:
https://www.nytimes.com/interactive/2018/09/19/upshot/facebook-county-friendships.html?action=click&module=In%20Other%20News&pgtype=Homepage&action=click&module=News&pgtype=Homepage

Banque de France has been for many years at the forefront of disseminating statistical data to academics and other interested parties. Through Banque de France’s dedicated public portal http://webstat.banque-france.fr/en/, we offer a large set of free downloadable series (about 40 000 mainly aggregated series).

Banque de France has expanded further the service provided and launched, in Paris, in November 2016 an “Open Data Room”, providing researchers with a free access to granular data.

We are glad to announce that the “Open Data Room” service is now also available to US researchers through Banque de France Representative Office in New York City.

How could I have forgotten to announce my latest paper, "Machine Learning for Regularized Survey Forecast Combination: Partially-Egalitarian Lasso and its Derivatives"? (Actually a heavily-revised version of an earlier paper, including a new title.) Came out as an NBER w.p. a week or two ago.

Missing Events in Event Studies: Identifying the Effects of Partially-Measured News Surprises
by Refet S. Guerkaynak, Burcin Kisacikoglu, Jonathan H. Wright #25016 (AP ME)
http://papers.nber.org/papers/w25016?utm_campaign=ntw&utm_medium=email&utm_source=ntw

Colacito, Ric, Bridget Hoffmann, and Toan Phan (2018) “Temperature and growth: A panel
analysis of the United States,”
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2546456

Do You Know That I Know That You Know...? Higher-Order Beliefs in Survey Data
by Olivier Coibion, Yuriy Gorodnichenko, Saten Kumar, Jane Ryngaert #24987 (EFG ME)
http://papers.nber.org/papers/w24987?utm_campaign=ntw&utm_medium=email&utm_source=ntw

Check out Goolgle's new Datset Search.

Here's the description they issued today:

Making it easier to discover datasets
Natasha Noy
Research Scientist, Google AI
Published Sep 5, 2018

In today's world, scientists in many disciplines and a growing number of journalists live and breathe data. There are many thousands of data repositories on the web, providing access to millions of datasets; and local and national governments around the world publish their data as well. To enable easy access to this data, we launched Dataset Search, so that scientists, data journalists, data geeks, or anyone else can find the data required for their work and their stories, or simply to satisfy their intellectual curiosity.

Similar to how Google Scholar works, Dataset Search lets you find datasets wherever they’re hosted, whether it’s a publisher's site, a digital library, or an author's personal web page. To create Dataset search, we developed guidelines for dataset providers to describe their data in a way that Google (and other search engines) can better understand the content of their pages. These guidelines include  salient information about datasets: who created the dataset, when it was published, how the data was collected, what the terms are for using the data, etc. We then collect and link this information, analyze where different versions of the same dataset might be, and find publications that may be describing or discussing the dataset. Our approach is based on an open standard for describing this information (schema.org) and anybody who publishes data can describe their dataset this way. We encourage dataset providers, large and small, to adopt this common standard so that all datasets are part of this robust ecosystem.

In this new release, you can find references to most datasets in environmental and social sciences, as well as data from other disciplines including government data and data provided by news organizations, such as ProPublica. As more data repositories use the schema.org standard to describe their datasets, the variety and coverage of datasets that users will find in Dataset Search, will continue to grow.

Dataset Search works in multiple languages with support for additional languages coming soon. Simply enter what you are looking for and we will help guide you to the published dataset on the repository provider’s site.

For example, if you wanted to analyze daily weather records, you might try this query in Dataset Search:

[daily weather records]

You’ll see data from NASA and NOAA, as well as from academic repositories such as Harvard's Dataverse and Inter-university Consortium for Political and Social Research (ICPSR). Ed Kearns, Chief Data Officer at NOAA, is a strong supporter of this project and helped NOAA make many of their datasets searchable in this tool. “This type of search has long been the dream for many researchers in the open data and science communities” he said. “And for NOAA, whose mission includes the sharing of our data with others, this tool is key to making our data more accessible to an even wider community of users.”

This launch is one of a series of initiatives to bring datasets more prominently into our products. We recently made it easier to discover tabular data in Search, which uses this same metadata along with the linked tabular data to provide answers to queries directly in search results. While that initiative focused more on news organizations and data journalists, Dataset search can be useful to a much broader audience, whether you're looking for scientific data, government data, or data provided by news organizations.

A search tool like this one is only as good as the metadata that data publishers are willing to provide. We hope to see many of you use the open standards to describe your data, enabling our users to find the data that they are looking for. If you publish data and don't see it in the results, visit our instructions on our developers site which also includes a link to ask questions and provide feedback.

The role of time-series statistics / econometrics in climate analyses is expanding (e.g., here).  Related -- albeit focusing on shorter-term meteorological aspects rather than longer-term climatological aspects -- it's worth listening to Michael Lewis' latest, The Coming Storm.  (You have to listen rather than read, as it's only available as an audiobook, but it's only about two hours.)  It's a fascinating story, well researched and well told by Lewis, just as you'd expect.  There are lots of interesting insights on (1) the collection, use, and abuse of public weather data, including ongoing, ethically-dubious, and potentially life-destroying attempts to privatize public weather data for private gain, (2) the clear and massive improvements in weather forecasting in recent decades, (3) behavioral aspects of how best to communicate forecasts so people understand them, believe them, and take appropriate action before disaster strikes. 

The 25-year accumulation of evidence for long memory / fractional integration / self-similarity / scaling laws in financial asset return volatility continues unabated.  For the latest see this nice new paper from Bank of Portugal, in particular its key Table 6. Of course the interval estimates of the fractional integration parameter "d" are massively far from both 0 and 1 -- that's the well-known long memory. But what's new and interesting is the systematic difference in the intervals depending on whether one uses absolute or range-based volatility. The absolute d intervals tend to be completely below 1/2 (0<d<1/2 corresponds to covariance-stationary dynamics), whereas the range-based d intervals tend to include 1/2 (1/2<d<1 corresponds to mean-reverting but not covariance- stationary dynamics, due to infinite unconditional variance). 

Realized vol based on the range is less noisy than realized vol based on absolute returns. But least noisy of all, and not considered in the paper above, is realized vol calculated directly from high-frequency return data (HFD-vol), as done by numerous authors in recent decades. Interestingly, recent work for HFD-vol also reports d intervals that tend to poke above 1/2. See this earlier post.

Markus Pelger has a nice paper on factor modeling with time-varying loadings in high dimensions. There are many possible applications. He applies it to level-slope-curvature yield-curve models. 

For me another really interesting application would be measuring connectedness in financial markets, as a way of tracking systemic risk. The Diebold-Yilmaz (DY) connectedness framework is based on a high-dimensional VAR with time-varying coefficients, but not factor structure. An obvious alternative in financial markets, which we used to discuss a lot but never pursued, is factor structure with time-varying loadings, exactly in Pelger! 

It would seem, however, that any reasonable connectedness measure in a factor environment would need to be based not only time-varying loadings but also time-varying idiosynchratic shock variances, or more precisely a time-varying noise/signal ratio (e.g., in a 1-factor model, the ratio of the idiosyncratic shock variance to the factor innovation variance). That is, connectedness in factor environments is driven by BOTH the size of the loadings on the factor(s) AND the amount of variation in the data explained by the factor(s). Time-varying loadings don't really change anything if the factors are swamped by massive noise. 
Typically one might fix the factor innovation variance for identification, but allow for time-varying idiosyncratic shock variance in addition to time-varying factor loadings. It seems that Pelger's framework does allow for that. Crudely, and continuing the 1-factor example, consider y_t  =  lambda_t  f_t  +  e_t. His methods deliver estimates of the time series of loadings lambda_t and factor f_t, robust to heteroskedasticity in the idiosyncratic shock e_t. Then in a second step one could back out an estimate of the time series of e_t and fit a volatility model to it. 
Then the entire system would be estimated and one could calculate connectedness measures based, for example, on variance decompositions as in the DY framework. 

Non-seasonally-adjusted U.S. GDP is finally here, as of a few days ago. See this BEA slide deck, pp. 10, 14-15. For background see my earlier post here. Also click on "Seasonality" under "Browse by Topic" on the right. 

The deck also covers lots of other aspects of the BEA's important "2018 Comprehensive Update of the National Income and Product Accounts". The whole deck is fascinating, and surely worth a close examination.

I'm on record as being largely unimpressed by the contributions of neural nets (NN's) in economics thus far. In many economic environments the relevant non-linearities seem too weak and the signal/noise ratios too low for NN's to contribute much. 

The Gu-Kelly-Xiu paper that I mentioned earlier may change that. I mentioned their success in applying machine-learning methods to forecast equity risk premia out of sample. NN's, in particular, really shine. The paper is thoroughly and meticulously done. 

This is potentially a really big deal.

This is adapted from remarks read at a memorial service earlier this week:

Just got back from the NBER Summer Institute. Lots of good stuff happening in the Forecasting and Empirical Methods group. The program, with links to papers, is here.

There is little doubt that climate change -- tracking, assessment, and hopefully its eventual mitigation -- is the burning issue of our times. Perhaps surprisingly, time-series econometric methods have much to offer for weather and climatological modeling (e.g., here), and several econometric groups in the UK, Denmark, and elsewhere have been pushing the agenda forward.

Now the NYU Volatility Institute is firmly on board. A couple months ago I was at their most recent annual conference, "A Financial Approach to Climate Risk", but it somehow fell through the proverbial (blogging) cracks. The program is here, with links to many papers, slides, and videos. Two highlights, among many, were the presentations by Jim Stock (insights on the climate debate gleaned from econometric tools, slides here) and Bob Litterman (an asset-pricing perspective on the social cost of climate change, paper here). A fine initiative!

For obvious reasons Peter Christoffersen has been on my mind. Here's an example of how his influence extended in important ways. Hopefully it's also an entertaining and revealing story.

Everyone knows Peter's classic 1998 "Evaluating Interval Forecasts" paper, which was part of his Penn dissertation. The key insight was that correct conditional calibration requires not only that the 0-1 "hit sequence" of course have the right mean ((1-\(\alpha\)) for a nominal 1-\(\alpha\) percent interval), but also that it be iid (assuming 1-step-ahead forecasts). More precisely, it must be iid Bernoulli(1-\(\alpha\)).

Around the same time I naturally became interested in going all the way to density forecasts and managed to get some more students interested (Todd Gunther and Anthony Tay). Initially it seemed hopeless, as correct density forecast conditional calibration requires correct conditional calibration of all possible intervals that could be constructed from the density, of which there are uncountably infinitely many.

Then it hit us. Peter had effectively found the right notion of an optimal forecast error for interval forecasts. And just as optimal point forecast errors generally must be independent, so too must optimal interval forecast errors (the Christoffersen hit sequence). Both the point and interval versions are manifestations of "the golden rule of forecast evaluation": Errors from optimal forecasts can't be forecastable. The key to moving to density forecasts, then, would be to uncover the right notion of forecast error for a density forecast. That is, to uncover the function of the density forecast and realization that must be independent under correct conditional calibration. The answer turns out to be the Probability Integral Transform, \(PIT_t=\int_{-\infty}^{y_t} p_t(y_t)\), as discussed in Diebold, Gunther and Tay (1998), who show that correct density forecast conditional calibration implies \(PIT \sim iid U(0,1)\). 

The meta-result that emerges is coherent and beautiful: optimality of point, interval, and density forecasts implies, respectively, independence of forecast error, hit, and \(PIT\) sequences. The overarching point is that a large share of the last two-thirds of the three-part independence result -- not just the middle third -- is due to Peter. He not only cracked the interval forecast evaluation problem, but also supplied key ingredients for cracking the density forecast evaluation problem.

Wonderfully and appropriately, Peter's paper and ours were published together, indeed contiguously, in the International Economic Review. Each is one of the IER's ten most cited since its founding in 1960, but Peter's is clearly in the lead!