Slide 1 of 25 - European Central Bank · 2016-09-20 · Panel Discussion on Uses of Models at...
Transcript of Slide 1 of 25 - European Central Bank · 2016-09-20 · Panel Discussion on Uses of Models at...
![Page 1: Slide 1 of 25 - European Central Bank · 2016-09-20 · Panel Discussion on Uses of Models at Central Banks ECB Workshop on DSGE Models and Forecasting September 23, 2016 John Roberts](https://reader036.fdocuments.us/reader036/viewer/2022070718/5eddf49fad6a402d6669333d/html5/thumbnails/1.jpg)
Panel Discussion on Uses of Models at Central BanksECB Workshop on DSGE Models and Forecasting
September 23, 2016
John RobertsFederal Reserve Board of Governors
Slide 1 of 25
John Roberts Federal Reserve Board of Governors
ECB Workshop on Models and Forecasting Panel Discussion
![Page 2: Slide 1 of 25 - European Central Bank · 2016-09-20 · Panel Discussion on Uses of Models at Central Banks ECB Workshop on DSGE Models and Forecasting September 23, 2016 John Roberts](https://reader036.fdocuments.us/reader036/viewer/2022070718/5eddf49fad6a402d6669333d/html5/thumbnails/2.jpg)
How are models used at the Board of Governors?
• For forecasting• For alternative scenarios and policy simulations
Disclaimer: The views expressed are my own and are not necessarily shared by other members of the staff or by the Board of Governors.
Slide 1 of 24
John Roberts Federal Reserve Board of Governors
ECB Workshop on Models and Forecasting Panel Discussion
![Page 3: Slide 1 of 25 - European Central Bank · 2016-09-20 · Panel Discussion on Uses of Models at Central Banks ECB Workshop on DSGE Models and Forecasting September 23, 2016 John Roberts](https://reader036.fdocuments.us/reader036/viewer/2022070718/5eddf49fad6a402d6669333d/html5/thumbnails/3.jpg)
Key questions
1. What models are used at various horizons?2. Will DSGE models eventually improve enough to replace
semi‐structural models?3. How is nonstandard data used at the BoG?4. How to interpret the “regime‐based” approach?
Slide 2 of 24
John Roberts Federal Reserve Board of Governors
ECB Workshop on Models and Forecasting Panel Discussion
![Page 4: Slide 1 of 25 - European Central Bank · 2016-09-20 · Panel Discussion on Uses of Models at Central Banks ECB Workshop on DSGE Models and Forecasting September 23, 2016 John Roberts](https://reader036.fdocuments.us/reader036/viewer/2022070718/5eddf49fad6a402d6669333d/html5/thumbnails/4.jpg)
1. Forecasting at the Federal Reserve Board of Governors
•Approach is ultimately judgmental•Consult many models•No single “core” model
Slide 3 of 24
John Roberts Federal Reserve Board of Governors
ECB Workshop on Models and Forecasting Panel Discussion
![Page 5: Slide 1 of 25 - European Central Bank · 2016-09-20 · Panel Discussion on Uses of Models at Central Banks ECB Workshop on DSGE Models and Forecasting September 23, 2016 John Roberts](https://reader036.fdocuments.us/reader036/viewer/2022070718/5eddf49fad6a402d6669333d/html5/thumbnails/5.jpg)
Models used to aid judgmental forecast
1. Near‐term• Non‐structural, time series
2. Medium‐term• Reduced‐form and semi‐structural
3. Longer‐term• A model‐based approach• Relatively structural
Slide 4 of 24
John Roberts Federal Reserve Board of Governors
ECB Workshop on Models and Forecasting Panel Discussion
![Page 6: Slide 1 of 25 - European Central Bank · 2016-09-20 · Panel Discussion on Uses of Models at Central Banks ECB Workshop on DSGE Models and Forecasting September 23, 2016 John Roberts](https://reader036.fdocuments.us/reader036/viewer/2022070718/5eddf49fad6a402d6669333d/html5/thumbnails/6.jpg)
1.1 Near‐term models
•Bottom‐up, decentralized approach• Example: GDP • Now‐casting attempts to replicate BEA (statistical agency) procedure• Decentralized specialist forecasts of spending components
• Factor models provide a cross‐check• In practice, given a low weight
Slide 5 of 24
John Roberts Federal Reserve Board of Governors
ECB Workshop on Models and Forecasting Panel Discussion
![Page 7: Slide 1 of 25 - European Central Bank · 2016-09-20 · Panel Discussion on Uses of Models at Central Banks ECB Workshop on DSGE Models and Forecasting September 23, 2016 John Roberts](https://reader036.fdocuments.us/reader036/viewer/2022070718/5eddf49fad6a402d6669333d/html5/thumbnails/7.jpg)
Decentralized approach leads to a very diverse set of models• Labor‐market flows
• Exploiting persistence in flows to predict unemployment rate• Residential construction
• Permit issuance predicts housing starts• Housing starts predict residential construction
• Business equipment spending• Two approaches:
1. Order books of capital‐goods producers (adjusted for exports and imports)
2. Reduced‐form models, focused on sentiment and financial conditions
Slide 6 of 24
John Roberts Federal Reserve Board of Governors
ECB Workshop on Models and Forecasting Panel Discussion
![Page 8: Slide 1 of 25 - European Central Bank · 2016-09-20 · Panel Discussion on Uses of Models at Central Banks ECB Workshop on DSGE Models and Forecasting September 23, 2016 John Roberts](https://reader036.fdocuments.us/reader036/viewer/2022070718/5eddf49fad6a402d6669333d/html5/thumbnails/8.jpg)
1.2 Medium‐term forecast
• Ultimately judgmental• Both bottom‐up and top‐down model input• Top‐down:
• Use of FRB/US and other multipliers to calibrate GDP responses to changes in financial and fiscal conditions
• Bottom up:• Decentralized• Reduced‐form and semi‐structural models used to inform judgment• Top‐down and bottom‐up tensions resolved through negotiation
Slide 7 of 24
John Roberts Federal Reserve Board of Governors
ECB Workshop on Models and Forecasting Panel Discussion
![Page 9: Slide 1 of 25 - European Central Bank · 2016-09-20 · Panel Discussion on Uses of Models at Central Banks ECB Workshop on DSGE Models and Forecasting September 23, 2016 John Roberts](https://reader036.fdocuments.us/reader036/viewer/2022070718/5eddf49fad6a402d6669333d/html5/thumbnails/9.jpg)
Examples of medium‐term models
• Okun’s Law• Linking GDP and labor‐market forecasts
• Reduced‐form consumption models• Relating consumer spending to income, financial conditions, consumer sentiment, and other factors
• Semi‐structural models of investment• Relating investment to business output and the user cost of capital• “Accelerator”
• Reduced‐form Phillips‐curve models of inflation
Slide 8 of 24
John Roberts Federal Reserve Board of Governors
ECB Workshop on Models and Forecasting Panel Discussion
![Page 10: Slide 1 of 25 - European Central Bank · 2016-09-20 · Panel Discussion on Uses of Models at Central Banks ECB Workshop on DSGE Models and Forecasting September 23, 2016 John Roberts](https://reader036.fdocuments.us/reader036/viewer/2022070718/5eddf49fad6a402d6669333d/html5/thumbnails/10.jpg)
1.3 The Long‐Term Forecast
•We recently adopted a simple semi‐structural model to generate our long‐term forecast.• Core is a New Keynesian “three‐equation model.”
• This model is close in spirit to a DSGE model.• It is calibrated to have properties similar to our main large‐scale semi‐structural model, FRB/US.
Slide 9 of 24
John Roberts Federal Reserve Board of Governors
ECB Workshop on Models and Forecasting Panel Discussion
![Page 11: Slide 1 of 25 - European Central Bank · 2016-09-20 · Panel Discussion on Uses of Models at Central Banks ECB Workshop on DSGE Models and Forecasting September 23, 2016 John Roberts](https://reader036.fdocuments.us/reader036/viewer/2022070718/5eddf49fad6a402d6669333d/html5/thumbnails/11.jpg)
Experience with our long‐term model
• In use since beginning of 2016.• Difficult to evaluate forecast performance of a long‐term model.• In our view, simple structure is a virtue:Very easy to use and explain.
Slide 10 of 24
John Roberts Federal Reserve Board of Governors
ECB Workshop on Models and Forecasting Panel Discussion
![Page 12: Slide 1 of 25 - European Central Bank · 2016-09-20 · Panel Discussion on Uses of Models at Central Banks ECB Workshop on DSGE Models and Forecasting September 23, 2016 John Roberts](https://reader036.fdocuments.us/reader036/viewer/2022070718/5eddf49fad6a402d6669333d/html5/thumbnails/12.jpg)
2. Will DSGE models eventually improve enough to replace semi‐structural models?
DSGE models can be assessed on two key dimensions:
1. Forecasting
2. Story telling
Slide 11 of 24
John Roberts Federal Reserve Board of Governors
ECB Workshop on Models and Forecasting Panel Discussion
![Page 13: Slide 1 of 25 - European Central Bank · 2016-09-20 · Panel Discussion on Uses of Models at Central Banks ECB Workshop on DSGE Models and Forecasting September 23, 2016 John Roberts](https://reader036.fdocuments.us/reader036/viewer/2022070718/5eddf49fad6a402d6669333d/html5/thumbnails/13.jpg)
2.1 EDO’s forecast performance• EDO is the Board staff’s oldest domestically oriented DSGE model: In use since 2009.•As a reminder, the Board staff doesn’t use EDO—or any other one model—to generate its baseline forecast.• It is nonetheless of interest to compare EDO’s real‐time forecasting performance with that of the Survey of Professional Forecasters (SPF).
Slide 12 of 24
John Roberts Federal Reserve Board of Governors
ECB Workshop on Models and Forecasting Panel Discussion
![Page 14: Slide 1 of 25 - European Central Bank · 2016-09-20 · Panel Discussion on Uses of Models at Central Banks ECB Workshop on DSGE Models and Forecasting September 23, 2016 John Roberts](https://reader036.fdocuments.us/reader036/viewer/2022070718/5eddf49fad6a402d6669333d/html5/thumbnails/14.jpg)
EDO (red) vs. SPF (blue)GDP forecasts
Slide 13 of 24
John Roberts Federal Reserve Board of Governors
ECB Workshop on Models and Forecasting Panel Discussion
![Page 15: Slide 1 of 25 - European Central Bank · 2016-09-20 · Panel Discussion on Uses of Models at Central Banks ECB Workshop on DSGE Models and Forecasting September 23, 2016 John Roberts](https://reader036.fdocuments.us/reader036/viewer/2022070718/5eddf49fad6a402d6669333d/html5/thumbnails/15.jpg)
EDO has performed somewhat less well
• Four‐quarter‐ahead RMSE is about 12 percent higher than the SPF.
But• SPF is an average over many forecasts—an advantage in forecasting.• A disadvantage in story‐telling.
Slide 14 of 24
John Roberts Federal Reserve Board of Governors
ECB Workshop on Models and Forecasting Panel Discussion
![Page 16: Slide 1 of 25 - European Central Bank · 2016-09-20 · Panel Discussion on Uses of Models at Central Banks ECB Workshop on DSGE Models and Forecasting September 23, 2016 John Roberts](https://reader036.fdocuments.us/reader036/viewer/2022070718/5eddf49fad6a402d6669333d/html5/thumbnails/16.jpg)
2.2 Models and stories
• In principle, DSGE models are well‐suited to story‐telling.• Because they are structural.
•Many practical policymakers disagree.Blanchard: DSGE models are bad communications devices.
Slide 15 of 24
John Roberts Federal Reserve Board of Governors
ECB Workshop on Models and Forecasting Panel Discussion
![Page 17: Slide 1 of 25 - European Central Bank · 2016-09-20 · Panel Discussion on Uses of Models at Central Banks ECB Workshop on DSGE Models and Forecasting September 23, 2016 John Roberts](https://reader036.fdocuments.us/reader036/viewer/2022070718/5eddf49fad6a402d6669333d/html5/thumbnails/17.jpg)
Models and stories, continued• Greater flexibility in model structure is an advantage of semi‐structural models.• Don’t need to wait for theory to catch up.
• Example: Wealth effect on labor supply.• Some more‐recent DSGE models have dropped it.
• There are lessons in both directions.• For example, in FRB/US, all dynamics are determined by adjustment costs. Would probably want to allow for serially correlated errors.
• Possible outcome: Eventual convergence.• Theoretical stories catch up with reality.• Better understanding of theoretical stories.
Slide 16 of 24
John Roberts Federal Reserve Board of Governors
ECB Workshop on Models and Forecasting Panel Discussion
![Page 18: Slide 1 of 25 - European Central Bank · 2016-09-20 · Panel Discussion on Uses of Models at Central Banks ECB Workshop on DSGE Models and Forecasting September 23, 2016 John Roberts](https://reader036.fdocuments.us/reader036/viewer/2022070718/5eddf49fad6a402d6669333d/html5/thumbnails/18.jpg)
2.3 How we use structural models at the FRBOG
• Not for forecasting• But we have long used our workhorse semi‐structural model, FRB/US, for many other purposes.• Alternative scenarios• Assessment of policy strategiesWe currently use DSGE models alongside FRB/US for these purposes.
• As just noted, we recently moved away from FRB/US, toward a simpler semi‐structural model, for our long‐term forecast.• Closer to a DSGE model
Slide 17 of 24
John Roberts Federal Reserve Board of Governors
ECB Workshop on Models and Forecasting Panel Discussion
![Page 19: Slide 1 of 25 - European Central Bank · 2016-09-20 · Panel Discussion on Uses of Models at Central Banks ECB Workshop on DSGE Models and Forecasting September 23, 2016 John Roberts](https://reader036.fdocuments.us/reader036/viewer/2022070718/5eddf49fad6a402d6669333d/html5/thumbnails/19.jpg)
3. Nonstandard data at the FRBOG
• Looking at a broad range of new data sources•Current view is that search data are not very helpful.• The experience of an early experiment in this area, Google Flu TrendsTop search result:
Slide 18 of 24
John Roberts Federal Reserve Board of Governors
ECB Workshop on Models and Forecasting Panel Discussion
![Page 20: Slide 1 of 25 - European Central Bank · 2016-09-20 · Panel Discussion on Uses of Models at Central Banks ECB Workshop on DSGE Models and Forecasting September 23, 2016 John Roberts](https://reader036.fdocuments.us/reader036/viewer/2022070718/5eddf49fad6a402d6669333d/html5/thumbnails/20.jpg)
Slide 19 of 24
John Roberts Federal Reserve Board of Governors
ECB Workshop on Models and Forecasting Panel Discussion
![Page 21: Slide 1 of 25 - European Central Bank · 2016-09-20 · Panel Discussion on Uses of Models at Central Banks ECB Workshop on DSGE Models and Forecasting September 23, 2016 John Roberts](https://reader036.fdocuments.us/reader036/viewer/2022070718/5eddf49fad6a402d6669333d/html5/thumbnails/21.jpg)
Nonstandard data, continued
•More promising are transactions data.•Credit‐card transactions•Payroll processing data
•Working with traditional statistical agencies to improve timeliness, coverage, sample sizes.• Evolutionary, not revolutionary
Slide 20 of 24
John Roberts Federal Reserve Board of Governors
ECB Workshop on Models and Forecasting Panel Discussion
![Page 22: Slide 1 of 25 - European Central Bank · 2016-09-20 · Panel Discussion on Uses of Models at Central Banks ECB Workshop on DSGE Models and Forecasting September 23, 2016 John Roberts](https://reader036.fdocuments.us/reader036/viewer/2022070718/5eddf49fad6a402d6669333d/html5/thumbnails/22.jpg)
4. The regime‐based approach
• Forecasting approach recently adopted by the St. Louis Fed•While the St. Louis forecast is formally judgmental, it has many points of contact with a Markov‐switching approach.• Focus on infrequent changes in regime.
•Will therefore consider some issues with forecasts in a formal Markov‐switching approach
Slide 21 of 24
John Roberts Federal Reserve Board of Governors
ECB Workshop on Models and Forecasting Panel Discussion
![Page 23: Slide 1 of 25 - European Central Bank · 2016-09-20 · Panel Discussion on Uses of Models at Central Banks ECB Workshop on DSGE Models and Forecasting September 23, 2016 John Roberts](https://reader036.fdocuments.us/reader036/viewer/2022070718/5eddf49fad6a402d6669333d/html5/thumbnails/23.jpg)
Forecasting in the face of nonlinearities
• Casually, Board staff would characterize their forecast as a mode rather than a mean.• Prompted by desire to have a narrative.• Brexit either happens or it doesn’t.
• In linear models with normal errors, the mean and the mode are the same, so there is no tension. • But (U.S.) business cycle appears to be nonlinear.
• Change in the unemployment rate is highly skewed.
Slide 22 of 24
John Roberts Federal Reserve Board of Governors
ECB Workshop on Models and Forecasting Panel Discussion
![Page 24: Slide 1 of 25 - European Central Bank · 2016-09-20 · Panel Discussion on Uses of Models at Central Banks ECB Workshop on DSGE Models and Forecasting September 23, 2016 John Roberts](https://reader036.fdocuments.us/reader036/viewer/2022070718/5eddf49fad6a402d6669333d/html5/thumbnails/24.jpg)
Nonlinearities, continued
• One useful nonlinear model is the Markov‐switching model.• In a two‐state Markov‐switching model, mean and modal forecasts can differ substantially.• If the states are persistent, the modal forecast will initially involve staying in one of the states.
• The mean forecast will instead be a weighted average of the two states.
• After a while, however, the probability of staying in the initial state becomes very low.
Slide 23 of 24
John Roberts Federal Reserve Board of Governors
ECB Workshop on Models and Forecasting Panel Discussion
![Page 25: Slide 1 of 25 - European Central Bank · 2016-09-20 · Panel Discussion on Uses of Models at Central Banks ECB Workshop on DSGE Models and Forecasting September 23, 2016 John Roberts](https://reader036.fdocuments.us/reader036/viewer/2022070718/5eddf49fad6a402d6669333d/html5/thumbnails/25.jpg)
Preliminary assessment of regime approach
•Promising, because it recognizes the world isn’t linear.•Unresolved issues about interpretation of longer‐horizon forecasts.
Slide 24 of 24
John Roberts Federal Reserve Board of Governors
ECB Workshop on Models and Forecasting Panel Discussion