Tuesday, June 30, 2020

Entering (and Trying to Exit) the Pandemic Recession

New paper at arXiv:2006.15183 
econ.EM econ.GN
Real-Time Real Economic Activity: Exiting the Great Recession and Entering the Pandemic Recession
AbstractWe study the real-time signals provided by the Aruoba-Diebold-Scotti Index of Business conditions (ADS) for tracking economic activity at high frequency. We start with exit from the Great Recession, comparing the evolution of real-time vintage beliefs to a "final" late-vintage chronology. We then consider entry into the Pandemic Recession, again tracking the evolution of real-time vintage beliefs. ADS swings widely as its underlying economic indicators swing widely, but the emerging ADS path as of this writing (late June) indicates a return to growth in May. The trajectory of the nascent recovery, however, is massively uncertain, particularly as COVID-19 spreads in the South and West, and could be reversed as quickly as it started.
Submitted 26 June, 2020; originally announced June 2020.

Monday, June 29, 2020

Causality and Generalized Impulse Response Functions

Neil Shephard just gave a fine talk, "Econometric analysis of potential outcomes time series:
instruments, shocks, linearity and the causal response function".  Recording here (soon). 
Slides here.  The key result is on slide 11:  If the conditions (Assns 1-3 on slides p. 5) for a potential outcome time series are satisfied, then the Koop-Pesaran-Potter (1996) "generalized impulse response function" (GIRF) has a direct causal interpretation. Neil pitched the paper as providing deeper understanding and firmer foundations for the GIRF, which it certainly does.

Wow! This is wonderful in general, and for me personally:  Throughout almost all my work with Kamil Yilmaz on measuring network connectedness (e.g., here), we work in a GIRF framework for the underlying vector autoregression (actually generalized variance decomposition, but it's the same thing). We liked the GIRF for a pragmatic reason -- its invariance to variable ordering, unlike Cholesky factor identification -- but we always wanted to understand it more deeply.

Thursday, June 25, 2020

Pandemic Economic Forecasting with Mixed-Frequency Data

Nice Schorfheide-Song (SS) Bayesian real-time pandemic economic forecasting with mixed-frequency data here.  They simulate exact posteriors.  ADS nowcasting is also based on exact mixed-frequency estimation (MLE).  So both SS and ADS get things right in principle (SS Bayesian forecasting, ADS frequentist nowcasting), and both are easily implemented in practice.  That is, both are intellectually pure, yet practically relevant.

Monday, June 22, 2020

COVID, Economic Activity, and Climate

Everyone talks about COVID helping reduce warming:
COVID up --> economic activity down --> CO2 down --> temperature down.

But there's a flip side:
COVID up --> activity down --> atmospheric sulphate aerosols down --> temperature UP!
(Sulphate aerosols reflect solar heat, so if they're down, temp is up.

See https://news.mongabay.com/2020/06/climate-conundrum-could-covid-19-be-linked-to-early-arctic-ice-melt/.

It would be interesting to assess the the competing effects of CO2 vs. sulphate aerosols, dynamically.  One might start with impulse-response analysis of an economic activity shock in a predictive model containing economic activity, CO2, sulphate aerosols, and temperature.

Wednesday, June 17, 2020

Time Series Modeling of COVID-19 Paths

Check out the refreshing new paper by Andrew Harvey and Paul Kattuman "Time series models based on growth curves with applications to forecasting coronavirus", pp. 126-156 here.

For non-causal forecasting, reduced-form approaches like those of Harvey-Kattuman are almost always the way to go, from traditional time series modeling to more recent extensions in machine learning. To paraphrase a long-ago No Hesitations post:  We generally don't need deep structural understanding to succeed at forecasting, which is wonderful, because we typically don't have deep structural understanding. (Admit it.) 

Forecasting COVID progression (cases, deaths, etc.) is a fine example. The leading structural ("SIR") model is a toy model, an intentionally stripped-down abstraction of a much more complex reality.  There's nothing wrong with that -- that's what all structural models are, and good structural models can yield invaluable insights. But good forecasting requires capturing the complex reality more fully, with its model uncertainty, measurement uncertainty, parameter uncertainty, innovation uncertainty, structural change uncertainty, etc. That's where reduced-form approaches shine.

On the other hand, because structural models can in principle illuminate the causal mechanisms that underlie reduced-form correlations, they may help with analysis of conterfactuals. That is, structural models may facilitate causal forecasting in addition to non-causal forecasting.

Of course it doesn't have to be an either/or choice.  One can attempt to blend the structural and reduced-form approaches, hoping to achieve the best of both worlds. To that end, see the also-refreshing new paper by Andrew Atkeson et al., "Estimating and forecasting disease scenarios for COVID-19 with an SIR model", here.  

Tuesday, June 16, 2020

SoFiE 2021 San Diego and 2022 Cambridge

Happy to help spread the word that the Society for Financial Econometrics annual conference 2021 will be in San Diego (UCSD) June 15-17, and 2022 will be in Cambridge England (University of Cambridge) June 27-29.  Finally some real in-person meetings, and each location is perfect.  Quite a big deal.  Zoom is hardly a substitute. See you there!

Monday, June 15, 2020

Did the U.S. Recession Start in February or March?

Some seem to think that the NBER declared a February recession start. It did not; rather, it declared a February cyclical peak.

So when did the recession start? It's a bit ambiguous, since a peak is an apex atop both the upward expansion path and the downward contraction path. Indeed the NBER's press release states that "The [February 2020] peak marks the end of the expansion that began in June 2009 and the beginning of a recession."

Nevertheless, when measuring expansion and contraction durations, the NBER convention is that peak months are taken as part of expansions ("the last month of good times"), and trough months are taken as part of contractions ("the last month of bad times"), as in the NBER table here.

Friday, June 12, 2020

Real Economics in Business Strategy Simulation

This spam somehow made it through my filters.  But it looks pretty cool, not really spam.  Maybe my filters are smarter than I thought.

https://scientificstrategy.com

I am certainly not expert in micro / IO / marketing, so maybe I'm far behind the curve, but at any rate the practitioner in me was intrigued by the tools described in the email below.  I have no idea whether they're any good, but it's certainly interesting to see real economics evidently getting in closer touch with the nitty-gritty of practical business decision making.

Hi Professor Diebold:

Model the dynamics of your market with Market Simulation:

Our 100+ example models include:

  • Cournot / Bertrand / Edgeworth / Giffen / Hotelling / Nash
  • Stackelberg Leader-Follower Price Competition
  • Wholesaler-Retailer Double Marginalization / M&A
  • eCommerce / Brick & Mortar
  • Good / Better / Best Product Pricing
  • Learning Curves / Search Costs / Bundling
  • Capacity Limitations / Switching Costs / Cannibalization
  • Conjoint Analysis / New Product Development

Our case studies include:

  • Android vs iOS 
  • Microbrews (6-parts)
  • Cola Wars (7-parts)
  • SUV Market (2-parts)
  • Competitive Strategy Game CSG (2-parts)
  • Porter’s Five Forces (5-parts)
    
Model, analyze, and solve your pricing / product / positioning / placement. Or send in your problem for us to solve.

Happy Simulations!

Ted.

-
Ted Hartnell | CTO
Phone: +1-415-800-4454
Address: 25 Pond Court, Milpitas, CA 95035

More on Conditional Predictive Accuracy Assessment

I recently blogged on the new Zhu-Timmermann paper.  I mentioned that they end on a constructive note for unconditional predictive accuracy comparisons, even if they raise issues for conditional comparisons.  I forgot (until now) about Li, Liao, and Quaedvlieg (2020), one of my favorite recent papers.  (We discussed it at length in my Ph.D. class in April.)  Their setup avoids the Zhu-Timmermann critique and provides an appealing route forward for conditional assessments.



Thursday, June 11, 2020

Monthly vs. Quarterly Recession Dating

The peak delineating the U.S. Pandemic Recession is Feb 2020 if you measure monthly (sure), but 2019Q4 if you measure quarterly (huh?).  See the NBER's explanation below.  Actually I find the explanation compelling: if you must measure quarterly, 2019Q4 is the right date.

The broader question is why look at a quarterly chronology when you have monthly? You might argue that our science is inexact, so that a monthly chronology may convey a false appearance of precision, like reporting econometric parameter estimates to ten decimal places.  Surely I have some sympathy for that view.

But I think the monthly chronology is generally reliable, and there is no doubt that, if I could have access only to monthly or quarterly, I'd take monthly.  Nevertheless quarterly is a useful complement, because you DON'T determine the peak quarter simply as the one containing the peak month, as the recent episode illustrates, so the quarterly chronology contains information not in the monthly chronology. The (fairly rare) situations when the peak quarter does not contain the peak month provide useful flags for situations deserving extra thought.

From NBER:

Determination of the February 2020 Peak in US Economic Activity

This report is also available as a PDF.
Cambridge, June 8, 2020 - The Business Cycle Dating Committee of the National Bureau of Economic Research maintains a chronology of the peaks and troughs of U.S. business cycles. The committee has determined that a peak in monthly economic activity occurred in the U.S. economy in February 2020. The peak marks the end of the expansion that began in June 2009 and the beginning of a recession. The expansion lasted 128 months, the longest in the history of U.S. business cycles dating back to 1854. The previous record was held by the business expansion that lasted for 120 months from March 1991 to March 2001.
The committee also determined that a peak in quarterly economic activity occurred in 2019Q4. Note that the monthly peak (February 2020) occurred in a different quarter (2020Q1) than the quarterly peak. The committee determined these peak dates in accord with its long-standing policy of identifying the months and quarters of peak activity separately, without requiring that the monthly peak lie in the same quarter as the quarterly peak. Further comments on the difference between the quarterly and monthly dates are provided below.
A recession is a significant decline in economic activity spread across the economy, normally visible in production, employment, and other indicators. A recession begins when the economy reaches a peak of economic activity and ends when the economy reaches its trough. Between trough and peak, the economy is in an expansion.
Because a recession is a broad contraction of the economy, not confined to one sector, the committee emphasizes economy-wide indicators of economic activity. The committee believes that domestic production and employment are the primary conceptual measures of economic activity.
The Month of the Peak
In determining the date of the monthly peak, the committee considers a number of indicators of employment and production. The committee normally views the payroll employment measure, which is based on a large survey of employers, as the most reliable comprehensive estimate of employment. This series reached a clear peak in February. The committee recognized that this survey was affected by special circumstances associated with the pandemic of early 2020. In the survey, individuals who are paid but not at work are counted as employed, even though they are not in fact working or producing. Workers on paid furlough, who became more numerous during the pandemic, thus resulted in an overcount of people working in recent months. Accordingly, the committee also considered the employment measure from the Bureau of Labor Statistics household survey, which excludes individuals who are paid but on furlough. This series plateaued from December 2019 through February 2020, and then fell steeply from February to March. Because both series measure employment during the week or pay period containing the 12th of the month, they understate the collapse of employment during the second half of March, as indicated by unprecedented levels of new claims for unemployment insurance. The committee concluded that both employment series were thus consistent with a business cycle peak in February.

The committee believes that the two most reliable comprehensive estimates of aggregate production are the quarterly estimates of real Gross Domestic Product (GDP) and of real Gross Domestic Income (GDI), both produced by the Bureau of Economic Analysis (BEA). These measures estimate production that occurred over an entire quarter and are not available monthly. The most comprehensive monthly measure of aggregate expenditures, which includes roughly 70 percent of real GDP, is monthly real personal consumption expenditures (PCE), published by the BEA. This series reached a clear peak in February 2020. The most comprehensive monthly measure of aggregate real income is real personal income less transfers, from the BEA. The deduction of transfers is necessary because transfers are included in personal income but do not arise from production. This measure also reached a well-defined peak in February 2020.
The Quarter of the Peak
In dating the quarterly peak, the committee relies on real GDP and real GDI as published by the BEA, and on quarterly averages of key monthly indicators. Quarterly real GDP and real GDI peaked in 2019Q4.

The quarterly average of employment as measured by the payroll series rose from 2019Q4 to 2020Q1. However, the committee concluded that the special factor noted above implies that the series should not play a significant role in determining the quarterly peak. The quarterly average as measured by the household survey reached a clear peak in 2019Q4. The committee concluded that like GDP and GDI, the number of people working also reached its quarterly peak in 2019Q4.

The fact that the monthly peak of February occurred in the middle of 2020Q1 while the quarterly peak occurred in 2019Q4 reflects the unusual nature of this recession. The economy contracted so sharply in March (the final month of the quarter) that in 2020Q1, GDP, GDI, and employment were significantly below their levels of 2019Q4.
Further Comments
The usual definition of a recession involves “a decline in economic activity that lasts more than a few months.” However, in deciding whether to identify a recession, the committee weighs the depth of the contraction, its duration, and whether economic activity declined broadly across the economy (the diffusion of the downturn). The committee recognizes that the pandemic and the public health response have resulted in a downturn with different characteristics and dynamics than prior recessions. Nonetheless, it concluded that the unprecedented magnitude of the decline in employment and production, and its broad reach across the entire economy, warrants the designation of this episode as a recession, even if it turns out to be briefer than earlier contractions.

Committee members participating in the decision were: Robert Hall, Stanford University (chair); Robert Gordon, Northwestern University; James Poterba, MIT and NBER President; Valerie Ramey, University of California, San Diego; Christina Romer, University of California, Berkeley; David Romer, University of California, Berkeley; James Stock, Harvard University; Mark Watson, Princeton University.

Wednesday, June 10, 2020

David Hendry on Graphene Nanotubes (!)

Penetrating climate insights  as always from David Hendry, this time on the potential role of graphene nanotubes in de-carbonizing.  https://voxeu.org/article/decarbonising-future-uk-economy.  I especially liked the bit about mandating that electric cars be plugged in when idle, supplying massive grid storage.  Not to mention electric planes.  Plus I can impress my friends at cocktail parties by tossing out "graphene nanotubes" (at least in a dream world that had cocktail parties).  More good Hendry & climate econometrics at http://www.climateeconometrics.org/ .


On Conditional Predictive Accuracy Comparisons

Zhu and Timmermann have a striking new paper, https://arxiv.org/abs/2006.03238, in which they show that the null hypothesis of equal conditional expected loss is logically impossible in the popular Giacomini-White environment.

So many fascinating  twists and turns in the forecast evaluation literature.  Unexpected issues continue to arise.

I like that the authors end on a positive note, with a constructive proposal for unconditional comparisons. even as they raise serious issues for conditional comparisons 

Tuesday, June 9, 2020

What Should We Name This Recession?

Too bad that "Great Recession" is already taken, ominous capitals and all, as it looks comically quaint compared to this one. So, what should we name this one?  Don't say "Great Great Recession".

I like "Pandemic Recession". It features not only well-deserved ominous capitals, but also clear provision of historical/causal color.  That's what I'm using.

Fastest-Ever NBER Recession Dating

Yesterday the NBER announced that the U.S. Pandemic Recession started in February 2020.  The four-month announcement lag is the shortest since 1980, and probably the shortest of all time. (See the table below, from work in progress. The announcement dates are in parentheses.) The short lag is noteworthy because typical NBER announcement lags are famously long, as is appropriate when establishing the scientific business-cycle chronology of record -- the NBER dating committee only gets one chance, and it wants to get it right. Indeed some might feel that yesterday's release was hasty! But I think not. They got it right, short announcement lag and all. Sometimes things come into sharp focus very quickly.  

Table: NBER Recessions

Recession Dates                                             Recession Characteristics






Starting Month
Ending Month
Duration
Depth
Severity
January 1980 (6/3/1980)
July 1980 (7/8/1981)
6
-3.6
-394.9
July 1981 (1/6/1982)
  November 1982 (7/8/1983)
 16
-2.9
-741.3
July 1990 (4/25/1991)
  March 1991 (12/22/1992)
8
-1.7
-376.8
March 2001 (11/26/2001)
November 2001 (7/17/2003)
8
-1.5
-328.4
December 2007 (12/1/2008)
June 2009 (9/20/2010)
18
-4.3
-1254.9
February 2020 (6/6/2020)
  ?
?
?
?


Saturday, June 6, 2020

SoFiE YouTube Channel & Conditional CAPM

The conditional CAPM is alive and well; now at ultra-high (intra-day) frequencies.  Witness, for example, the first SoFiE YouTube "show" (seminar), last week: "Recalcitrant Betas: Intraday Variation in the Cross-Sectional Dispersion of Systematic Risk and Expected Returns," by Torben G. Andersen, Martin Thyrsgaard, and Viktor Todorov,

https://www.youtube.com/channel/UCDUUkbiY_UcadeG7CeVCf2w/featured?view_as=subscriber

Things have come a long way since early GARCH work like

http://econ.duke.edu/~boller/Published_Papers/jpe_88.pdf

and early realized vol work like

https://www.sas.upenn.edu/~fdiebold/papers/paper59/RealizedBeta.pdf.

For the full SoFie seminar schedule see

https://sofie.stern.nyu.edu/node/6137/.

The seminar leader, Andrew Patton, runs a nice tight ship:  40 min presentation, 10 min discussant, 10 min Q&A keeps everyone focused and engaged.  And all seminars are recorded for later viewing.

Wednesday, June 3, 2020

Fine New Macro-Econometrics Toolbox

Check out the cutting-edge Ferroni and Canova toolbox and related materials, here.

From the abstract of their User's Guide:
We use MATLAB functions and routines to estimate VARs, FAVARs, local projections and other models with classical or Bayesian methods. The toolbox allows a researcher to conduct inference under various prior assumptions on the parameters, to produce point and density forecasts, and to trace out the causal effect of shocks using a number of identification schemes. The toolbox is equipped to handle missing observations. It can also deal with panels of time series. We describe the methodology employed and implementation of the functions. We illustrate the main features with a number of practical examples.
Looking forward to R, Python, and Julia versions!

Monday, June 1, 2020

Yield Curve Construction

An interesting new paper by Liu and Wu, https://www.nber.org/papers/w27266? and below, is the latest in traditional two-step yield curve construction (summarize bond prices with synthetic zero-coupon yields, and then treat the zero-coupon yields as data).  I wonder how their results would compare to the one-step approach of Andreasen, Christensen, and Rudebusch (2019), which directly analyzes the universe of bond prices,  https://www.sciencedirect.com/science/article/abs/pii/S0304407619300740 and below.  The one-step approach seems highly appealing. Liu and Wu may have built a better mousetrap, but Andreasen, Christensen, and Rudebusch arguably dispense with the need for a mousetrap.


Reconstructing the Yield Curve

Yan LiuJing Cynthia Wu

NBER Working Paper No. 27266
Issued in May 2020
NBER Program(s):Asset PricingEconomic Fluctuations and GrowthMonetary Economics
The constant-maturity zero-coupon Treasury yield curve is one of the most studied datasets. We construct a new dataset using a non-parametric kernel-smoothing method with a novel adaptive bandwidth specifically designed to fit the Treasury yield curve. Our curve is globally smooth while still capturing important local variation. Economically, we show that applying our data leads to different conclusions from using the leading alternative data of Gurkaynak et al. (2007) (GSW) when we repeat two popular studies of Cochrane and Piazzesi (2005) and Giglio and Kelly (2018). Statistically, we show our dataset preserves information in the raw data and has much smaller pricing errors than GSW. Our new yield curve is maintained and updated online, complemented by bandwidths that summarize information content in the raw data: https://sites.google.com/view/jingcynthiawu/yield-data.

Term Structure Analysis with Big Data: One-Step Estimation Using Bond Prices







Abstract

Nearly all studies that analyze the term structure of interest rates take a two-step approach. First, actual bond prices are summarized by interpolated synthetic zero-coupon yields, and second, some of these yields are used as the source data for further empirical examination. In contrast, we consider the advantages of a one-step approach that directly analyzes the universe of bond prices. To illustrate the feasibility and desirability of the one-step approach, we compare arbitrage-free dynamic term structure models estimated using both approaches. We also provide a simulation study showing that a one-step approach can extract the information in large panels of bond prices and avoid any arbitrary noise introduced from a first-stage interpolation of yields.