Quantitative methods for economic policy:

limits and new directions

Ignazio Visco

Banca d’Italia

Philadelphia, 25 October 2014

Outline

I. Before the outbreak of the global financial crisis

II. Limits unveiled1. Real-financial linkages2. Non-linearities3. Increased interconnectedness

III. Quantitative challenges for macroeconomic policy

4. Taking advantage of large datasets5. Modeling inflation expectations 6. Identifying structural vs. cyclical

developments7. Macroprudential policy

2

Before the outbreak of the global financial crisis Policymaking tools: from large-scale

macroeconometric models to more structural, medium-size “microfounded” DSGE models

Policy analysis framework in central banks: New Keynesian (NK) DSGE models o Rational expectations (RE), representative

agent, real/nominal rigiditieso Structural interpretation, complement to VAR

analysis, positive and normative use

Forecasting: large-scale modelso Flexibility, role of judgmento Provide detailed description of the economy

(pros and cons)3

4

Before the outbreak of the global financial crisis

Source: Banca d’Italia staff calculations *Obtained using a (non-centered) 10-year moving window

Before the outbreak of the global financial crisis

5

“macroeconomics in this original sense has succeeded: its central problem of depression prevention has been solved, for all practical purposes, and has in fact been solved for many decades”

Robert Lucas, 2003“the state of macro is good”

Olivier Blanchard, 2008

Before the outbreak of the global financial crisis

Financial resources collected by private sector

(percentage of GDP)

OTC and exchange-traded derivatives in US

(notional value, trillion of USD)

Source: Banca d’Italia staff calculations Source: Banca d’Italia staff calculations

6

Before the outbreak of the global financial crisis

“Philosophically, I do not believe that the market system, in even its purest form, provides adequate self-regulatory responses. The economy definitely needs guidance – even leadership – and it is up to professional economists to provide public policy makers with the right information to deliver such leadership. As for the methods of doing this, I see no alternative to the quantitative approach of econometrics, but I do realize that all policy issues are not quantitative and measurable. At times, subjective decisions must also be made.”

Lawrence Klein (1992)

7

The outbreak of the global financial crisis

Source: Chung, Laforte, Reifschneider, Williams (2012)

FRB/US Assessment of the Likelihood of Recent Events:

History Versus 2007Q4 Model Projection

8

The outbreak of the global financial crisis

Yet, explaining the dynamics of the crisis is crucial.

Analytical toolbox for macroeconomic policy must be repaired and updated

“One thing we are not going to have, now or ever, is a set of models that forecasts sudden falls in the value of financial assets, like the declines that followed the failure of Lehman Brothers”

Robert E. Lucas (2009)

“The crisis has made it clear that this view was wrong and that there is a need for a deep reassessment.”

Olivier Blanchard (2014) 9

Limits unveiled

1. Real-financial linkages

2. Non-linearities

3. Increased interconnectedness

10

Limit #1: Real-financial linkages

“If the real sector of the economy does not function so well, for instance, if it is dynamically unstable under some circumstances […] then the need for stabilization policies is hard to deny, and with it the need to model financial and monetary sectors of the economy”

Albert Ando (1979)

11

Limit #1: Real-financial linkages

No financial sector in pre-crisis, workhorse NK models used for policy analysis: one interest rate enough to track cyclical dynamics and support normative analysis

Why? Efficient Markets Hypothesis (EMH) behind the scenes: market clearing and RE guarantee that all information is efficiently used. No need to explicitly model financial sector…

…nonetheless, significant work on financial factors in pre-crisis NK models (e.g. financial accelerator)

Important (overlooked) contributions in macroeconomic literature: e.g. debt deflation, financial crises

12

Limit #1: Real-financial linkages

The crisis has ignited promising research in this area. Medium-scale NK models enriched along several dimensions:o inclusion of financial intermediation and

liquidityo private-sector leverage over the cycle and

role of institutions o modelling unconventional monetary

policy. Which channels? Liquidity, credit, expectations

Departures from representative agent framework

More attention to country-specific institutional features: shadow banking, sovereign risk, sovereign-banking linkages

Risk and uncertainty: rediscovery of Knightian uncertainty

13

Limit #1: Real-financial linkages

Large-scale macroeconometric models also shared the absence of significant real-financial interactions

However, they have historically proved to be flexible tools, open to non-mechanical use of external information (with “tender loving care”), especially in the occasion of unexpected breaks in empirical regularities

E.g. Klein (first oil shock, 1973): embed external information in the Wharton and LINK model to account for unprecedentedly large shock on oil prices, that no model could handle

In a similar vein today: role of credit in Bank of Italy model

14

Limit #1: Real-financial linkages

External information on loan supply restrictionsEffect on current-year GDP forecast error in 2008-2009 and

2011-2012 recessions

Source: Rodano, Siviero and Visco (2014) 15

Limit #2: Nonlinearities

Pre-crisis empirical models were best suited to deal with “regular” business cycles

The crisis marked a huge discontinuity with the past…

…in non-stationary environments, predictions based on past probability distributions can differ persistently from actual outcomes

Problems with existing models: o Not enough information within historical data

about shocks of such size and nature (“dummying out” of rare events)

o Linear dynamics cannot properly account for shock transmission and propagation

16

Limit #2: Nonlinearities

Advances in non-linear macroeconomic modeling

Models with time-varying parameters and stochastic volatilityo Flexible, although structural interpretation

may become tricky if all parameters are allowed to change

o Large shocks and non-Gaussian (tail) dependence: Can macro borrow from financial econometrics?

Regime-switching modelso Good in-sample fit. Less clear performance in

out-of-sample forecasting

Nonlinear methods in NK modelso Global methods account for occasionally

binding constraints, uncertainty and to go beyond “small” shocks. Which/how many nonlinearities?

17

Limit #3: Increased interconnectedness Trade linkages: (non-LINK) model forecasts

typically rely on assumptions about world demand, commodity prices, exchange rates (all exogenous variables). Open-economy dimension often contributes to large part of forecast errors, especially during crisis

Cross-border financial integration has markedly increased: need to go beyond trade linkages and account for foreign asset exposure, global banks

Methods: Global VAR, Panel-VARo Exploit cross-section data, static and

dynamic linkso Can account for changes in parameters that

capture cross-country linkages and spillovers

Applications of network theory to study interconnectedness

Modeling issues: common shocks or contagion?

18

Current challenges for macroeconomic policy

1. Taking advantage of large datasets

2. Modeling inflation expectations

3. Identifying structural vs. cyclical developments

4. Macroprudential policy

19

Challenge #1: Taking advantage of large datasets

“My present approach is to construct simple time-series models of high frequency data based on latest information, by days or weeks or months - for use in somewhat lower frequency macromodels […] I am a proponent of combining different sources of information, and the information source in this case is cross-section data from survey investigations. They should be integrated within macromodels.”

Lawrence Klein (1987)20

In times of crisis, the availability of accurate data is more crucial for policy analysis than it is in “normal” times o The more timely, accurate and relevant the

data, the better our assessment of the current state of economic activity

Various econometric instruments exploit data of different types and sources to produce good “nowcasts”o bridge models and MIDAS o large Bayesian VARso factor models (Banca d’Italia: €-Coin)

Combining evidence from models based on various datasets and assumptions (‘thick modeling’: Granger) as a way to account for growing uncertainty

Challenge #1: Taking advantage of large datasets

21

Challenge #1: Taking advantage of large datasets

“Good predictions have two requisites that are often hard to come by. First they require either a theoretical understanding of the phenomena to be predicted, as a basis for the prediction model, or phenomena that are sufficiently regular that they can be simply extrapolated. Since the latter condition is seldom satisfied by data about human affairs (or even by the weather), our predictions will generally be only as good as our theories. The second requisite for prediction is having reliable data about the initial conditions – the starting point from which the extrapolation is to be made.”

Herbert Simon (1972)

22

Source: Bank of Italy. For details see: Altissimo, F., Bassanetti, A., Cristadoro, R., Forni, M., Hallin, M., Lippi, M., Reichlin, L. and Veronese, G. (2001). A real Time Coincident Indicator for the euro area Business Cycle. CEPR Discussion Paper No. 3108; Altissimo, F., Cristadoro, R., Forni, M., Lippi, M., Veronese, G., New Eurocoin: Tracking economic growth in real time. The Review of Economics and Statistics, 2010

Challenge #1: Taking advantage of large datasets

€-coin indicator

Challenge #1: Taking advantage of large datasets

Nowcasting of many indicators can also benefit from use of ‘Big Data’: e.g. Google-based queries of unemployment benefits claims, car and housing sales, loan modification, etc.

Technological advances have made available a massive quantity of data, which offer potentially useful information for statistical and economic analysis (back, now and forecast)

Machine learning techniques: useful to cope with data of such size; can be applied to detect patterns and regularities, but… what role for economic theory?

“deep statistical theory with much stochastic structure in the analysis, […] is no substitute for economic theory. […] Without theory and other a priori information, we are lost”

Lawrence Klein (1977)24

Challenge #2: Modeling inflation expectations At the zero lower bound, repeated downward

revisions in inflation expectations may trigger a self-fulfilling deflationary spiral

Persistent differences in actual and expected inflation question the validity of the RE assumption in policy models

It is unlikely that households and firms can completely discount the effects of current and future policies in their demand and pricing decisions

Macromodels for policy analysis have largely ignored research on:o Learning mechanisms (example)o Rational inattentiono Behavioural economics

25

Challenge #2: Modeling inflation expectations

Inflation expectations and price stability in the euro area Rational expectations vs. adaptive learning

Source: Banca d’Italia; simulation of Clarida, Galí and Gertler 1999

26

Challenge #3: Structural vs. cyclical developments Financial crises are typically followed by a

much slower recovery than “normal” recessions (the current one is no exception)

For policy analysis it is imperative to disentangle the structural and cyclical effects of the Great Recession (although the two tend to be intimately related)o changes in “natural” rateso unemployment hysteresis effects

Large uncertainty surrounds global growth prospects

o “Secular stagnation” o “Second Machine Age”

How to design appropriate macroeconomic policies? E.g. fiscal policy…

27

Challenge #3: Structural vs. cyclical developments With the global financial crisis, public debt has

reached record peacetime levels in many advanced economies

High levels of public debt are a source of vulnerability and possible nonlinearities. How to measure fiscal sustainability and model its effects on sovereign risk?

Success of consolidation depends on credibility as well as on long-run structural measures to increase potential output

Models must account for both long and short-term factors

“There is nothing in the Keynesian prescriptions to support highly unbalanced policies or excessive reliance on monetary policy to provide economic stabilization”

Lawrence Klein (1992)

28

Challenge #4: Macroprudential policy

Macroprudential policies: maintain stability of financial system through containing systemic risks by increasing the resilience of the system and leaning against build-up of financial imbalances

What are the sources of financial cycles? o Financial shocks, news shocks, risk/uncertainty

shocks

What are the sources of systemic risk? o Pecuniary externalities, endogenous risk

What are the boundaries of the financial system? o Regulatory arbitrage, shadow banking system

How to assess conflicts and complementarity between monetary, micro and macroprudential policy?

29

Challenge #4: Macroprudential policy

Monitoring financial instabilityo Density forecasts and tail eventso Early warning: which models/variables?

Data: effort in identifying data needs (G20 Data Gaps Initiative)

Empirical evidence on macroprudential policy effectiveness:o So far mostly on EMEs (evidence not clear-cut)o Identification issues: macroprudential used in

conjunction with other policies Methods

o Event studies, stress tests, panel regressions, micro-data analysis, regime-switching, “microfounded ”.

o Suite of models?

30

Conclusion (I)

31

“The history of science, like the history of all human ideas, is a history of irresponsible dreams, of obstinacy, and of error. But science is one of the very few human activities — perhaps the only one — in which errors are systematically criticized and fairly often, in time, corrected. This is why we can say that, in science, we often learn from our mistakes, and why we can speak clearly and sensibly about making progress there.”

Karl Popper (1963)

Conclusion (II)

32

“It is my firm belief that the only satisfactory test of economics is the ability to predict, and in crucial predictive situations such as reconversion after World War II, the settlement of the Korean War, the settlement of the Vietnam War, the abrupt economic policy switch of the Nixon Administration in August 1972, the oil shock of 1973 (forecast of a world-wide succession by LINK), the recession of 1990. In these crucial periods, econometric models outperformed other approaches, yet there is considerable room for improvement, and that is precisely what is being examined in development of high-frequency models that aim to forecast the economy, every week, every fortnight, or every month, depending on the degree of fineness of the information flow.”

Lawrence Klein (2005)

Thanks

33