Show Summary Details

# (p. 691) Index

(p. 691) Index

“absorbing barrier,” 328

accuracy, forecast.

*See*forecast accuracyACD (autoregressive conditional duration) model, 518–19

ACR model, 336

ADL (autoregressive distributed lag)-MIDAS regressions, 231–35

administrative data sets, for health care cost models, 627

aggregating disaggregated forecasts, 291–92

AIM algorithm, 96

Akaike's information criterion.

*See*AICalgebra, of consensus forecasts, 464–67

analogy forecasting, 76

analytical point forecasts, 72–73

ANNs.

*See*neural networksapproximating logistic transition function, 330–31

approximators.

*See*universal approximatorsARCH (autoregressive conditional heteroskedasticity) models.

*See also*GARCH modelsforecast failure and, 280

HARCH, 541

MIDAS regressions and, 239

AR_FAC (multivariate factor-based models), 361.

*See also*forecast combinations illustration: survey and time series forecastsARFIMA (autoregressive fractionally integrated moving average) models, 526

ARMA structure and, 526–27

GARCH models

*v*., 540–41nonlinearity and, 543

white-noise process and, 514

ARIMA (autoregressive integrated moving average) models

random walk hypothesis and, 505

reduced-form volatility forecasting and, 527

sampling the future algorithm and, 102

structural time series models

*v*., 131ARMA (autoregressive moving average) models

ACD model and, 519

ADL-MIDAS regressions

*v*., 232Bass model and, 677

conditional predictability testing and, 453

cycle component as, 136

data revisions model and, 253

expert forecasts and, 686

multivariate version of, 151

smooth trend and, 139–41

UC-RV model and, 541

vector fractionally integrated, 541

Arrow, Kenneth, 561

Artificial neural networks.

*See*neural networksassessing forecast performance.

*See*forecast performance assessmentasymptotic approximation

with growing in-sample size, 443

autocorrelation functions.

*See*ACFsautomatic statistical procedure, nowcasting

*v*., 195autoregressive conditional duration ACD model, 518–19

autoregressive conditional heteroskedasticity.

*See*ARCH modelsautoregressive distributed lag MIDAS.

*See*ADL-MIDAS regressionsautoregressive fractionally integrated moving average.

*See*ARFIMA modelsautoregressive integrated moving average.

*See*ARIMA modelsautoregressive neural network models, 79

autoregressive representation, of unobserved components time series models, 148

Bachelier, Louis, 504–5

Finance Act of 1984, 321

Bank of England's Survey of External Forecasters, 467

Bass model, 674–82.

*See also*sales forecastingforecasting moment of peak sales with, 680–81

generalized, 682

important extensions of, 681–82

sales forecasts generated from, 676–78

Bayesian estimation, of dynamic factors, 44–46

Bayesian forecasting.

*See*sampling the future methodBayesian information criterion.

*See*BICBayesian reduced-rank VECMs, 18–19

Bayesian VARs.

*See*BVARsBayesian VECMs, 25

BCEI.

*See*Blue Chip Economic IndicatorsBFGS (Broyden-Fletcher-Goldfarb-Shannon) method, 143

biases (in judgment), 164, 167–72.

*See also*judgmentcontrarianism and, 167

egocentric attribution, 171

false consensus effect, 171

heuristics, 168–70

illusion of control, 171

“publicity hypothesis” and, 167

representativeness heuristic, 168–69

“reputational cheap talk” hypothesis and, 167

“wishful expectations hypothesis” and, 168

binary health care cost models, 626

“black arts,” 174

Board of Governors of the Federal Reserve, 195

breaks.

*See*structural breaks and location shiftsBroyden-Fletcher-Goldfarb-Shannon (BFGS) method, 143

business cycle component.

*See*cycle componentBVARs (Bayesian VARs).

*See also*NAWM benchmarksadvances in specification/estimation of, 9

DSGEs with, 104

VECMs and, 19

Capital asset pricing model(CAPM), 520

CAPM.

*See*capital asset pricing modelCaptain Cook/Maori example, 321

carbon emissions markets, 609.

*See also*electricity spot marketsenergy commodities bundle and, 609

EUAs and, 609

case studies, weather/climate forecasts, 570–73

CCA.

*See*canonical correlation analysisCenter for Economic Studies ifo Institute World Economic Survey, 467

CFNAI (Chicago Fed's National Activity Index), 430

chance, election forecasting and, 669

Chicago Fed index, 204

Chicago Fed's National Activity Index (CFNAI), 430

choice questions, 574

civil war example, 322

cleaning data, 532

climate forecasts.

*See*weather and climate forecastscoal market.

*See also*electricity spot marketscarbon emissions markets and, 609

cointegration analyses and, 608

volatility of, 620

Cochrane–Orcutt estimator, 42

cognitive feedback, 175

cointegration

CCA and, 10

coal market and, 608

common trends and, 263

DFMs and, 53–54

forecasting impacted by, 16–17

foreign exchange rates and, 521

fractional, 521

interest rates term structure and, 521

oil market and, 608

real business cycle theory and, 10

combination forecasting.

*See*forecast combinationscombining survey forecasts.

*See*survey forecastscommon nonlinear features, 17

conditional predictive ability tests, 4, 441–55

decision rules for forecast selection, 452–53

econometric methodology, 444–52

extensions, to different testing problems, 450–52

“fixed forecasting scheme” and, 442

illustration of, 441–42

open research questions, 453–54

time-varying predictive ability testing, 446–50

when to use, 443–44

Consensus Economics, Inc., 492

consensus forecasts, 457–58, 464–67.

*See also*survey forecastsaccuracy of, 464–65

algebra of, 464–67

anchoring and adjustment bias and, 170

arguments against, 469

“wisdom of crowds,” 465

contrarianism, 167

cost-effectiveness analysis, health care costs models and, 626

(p. 694)
cost-loss decision-making model, 560, 567–70, 571.

*See also*weather and climate forecastsdynamics in, 570

real-world situations and, 570

cost regression health care costs models, linear, 627–29

count data health care costs models, 626

cross-forecaster dispersion, in point predictions, 462–63

cross-sectional averaging estimators, 39–40

cumulative sum (CUSUM) test, 447

“curse of knowledge,” 172

CUSUM (cumulative sum) test, 447

cycle component, 129–30.

*See also*smooth cycle componentARMA process, 136

multiple cycles, 137

random walk plus smooth cycle model, 140–41

smooth time-varying trigonometric, 137

specification, 136–37

time-varying trigonometric, 136–37

data/data sets.

*See also*high-frequency data; information; MIDAS regressions; mixed-frequency data; multidimensional panel data of survey forecasts; panel-data sets; real-time dataadministrative, for health care cost models, 627

cleaning, 532

DFMs and, 35

large real-time databases (online list), 252

*Microeconometrics Using Stata*, 644

data-generating processes.

*See*DGPsdata issues.

*See also*missing data; mixed-frequency data; nowcasting; real-time data vintagesmodels and, 1

overview, 3

data revisions.

*See also*data revisions example; real-time dataDFMs and, 261–62

G7 countries (study), 254

location shifts and, 307–8

model of, 253–54

news revisions

*v*. noise revisions issue, 260–61optimal forecasting when data revisions exist, 259–64

real-time data analysis bibliography (online), 254

real-time data

*v*., 257–58RMAR and, 253

sizes of, 252–53

unconditional predictive ability tests and, 428–29

unresolved questions, 264

data revisions example (residential investment)

data snooping, 391–92

DCG (Diagnostic Cost Group) system, 626

DDD (double-differenced device), 283–84

disaggregation by time and, 301

disaggregation over variables and, 294–96

LDGP nonconstancy and, 284–87

VEqCM

*v*., 283–87decision analytic valuation studies, 560, 567–73.

*See also*weather and climate forecastscase studies, 570–73

limitations of, 579

prototypical decision-making models, 567–70

demography.

*See also*long-horizon economic growth forecastingGDP process and, 586

stable populations and, 589–90

DFMs.

*See*dynamic factor modelsDGPs (data-generating processes).

*See also*LDGPsreduced-rank VECMs and, 19–25

structural breaks and, 272

unconditional predictive ability Monte Carlo experiments and, 430–36

Diagnostic Cost Group (DCG) system, 626

diagnostic tests.

*See*health care costs models;*specific tests*disaggregation over variables, 273, 289–96, 309.

*See also*forecasting in presence of unanticipated location shiftsaggregating disaggregated forecasts, 291–92

break prediction and, 326

DDD and, 294–96

forecasting aggregate directly by its past, 292–94

with unanticipated breaks, 291

discrete-time GARCH models, 543

distributed lag (DL) models

MIDAS regression models and, 227–28

VARS models and, 10

volatility forecasting and, 537–40

DL models.

*See*distributed lag modelsdouble differenced device.

*See*DDDDSGEs (dynamic stochastic generale quilibrium models).

*See also*NAWMBVARs with, 104

increase in scale of, 123

misspecification and, 124

reduced-form models

*v.*, 90DSGEs, log-linearized, 89–124.

*See also*NAWM estimating predictive distribution of, 101–3, 123state-space form for, 90

dynamic factor models (DFMs), 2, 35–59.

*See also*dynamic factors; model instability illustration-DFMs combination; nowcasting model; static factorsADL-MIDAS models and, 233–35

appeal of, 54

cointegration and, 53–54

data revisions and, 261–62

data sets and, 35

early research, 35

error correction in, 53–54

extensions of, 52–54

forecast combinations and, 306–7

hierarchical, 54

motivation for using, 36

nonlinear, 45–46

nonlinear time series models

*v*., 2outlook for, 54–55

pseudo out-of-sample forecasting exercises and, 49

recent literature on, 36

with structural breaks, 52–53

SVARs and, 55

with time-varying parameters, 52–53

dynamic factors

Bayesian estimation of, 44–46

innovations, estimation of, 42

as instrumental variables, 50

in second-stage regressions, 48–50

state-space model with, 43–44

time-domain estimation of, 37–44

uses of, 48–52

dynamic principal components, 42

ECM (exponential conditional mean) health care costs models, 632–33, 634, 636, 639, 645, 646, 648

*t*, 649*t*econometric framework

conditional predictive ability testing, 444–52

econometric models, robust forecasting models

*v*., 279–88economic growth.

*See*long-horizon economic growth forecastingeconomic value of weather and climate forecasts.

*See*weather and climate forecastsECX (European Climate Exchange), 609

EEE models.

*See*extended estimating equations health care costs modelefficiency condition, Nordhaus's, 474–75

efficient capital markets theory, 505

egocentric attribution, 171

Einstein, Albert, 505

election forecasting statistical models

assessment, 663–66

generic model with examples, 660–62

reasons for selecting, 668

success of, 668–69

UK example, 662–63

electricity daily spot prices (Nord Pool exchange market), 130–31, 153–59

bivariate analysis, 156

forecasting results, 156–59

univariate analysis, 153–55

electricity spot prices

modeling with multiple explanatory drivers, 617–20

power spot modeling, 610

stochastic models of, 615–17

time series class of models and, 611

energy commodities (forecasting/modeling), 5, 607–21.

*See also*electricity spot marketsas bundle, 607–9

challenge of, 620

linkages of, 607–9

power market and, 620–21

error correction, in DFMs, 53–54

estimated factors.

*See*dynamic factorsEU emissions allowances (EUAs), 609

euro-area GDP.

*See*nowcasting modeleuro-area model.

*See*NAWMEurocoin index, 204

European Climate Exchange (ECX), 609

exact static factor model, 41

expectation maximization algorithm.

*See*EM algorithmexponential conditional mean health care costs models.

*See*ECM health care costs modelsexponential GAR CH.

*See*EGARCH modelextended estimating equations (EEE) health care costs model, 632, 639, 642, 642

*t*, 643, 645, 646, 647, 648*t*, 649, 649*t*factor error-correction model, 54

factor models.

*See*dynamic factor modelsfallowing/planting case study, 572

false consensus effect, 171

false discovery, 392

false discovery rate.

*See*FDRfamily-wise error rate.

*See*FWERFDR (false discovery rate)

controlling for, with given λ, 406–7

defined, 393

estimating, 405–6

optimal choice for λ, 407

Federal Open Market Committee, 484

feedback, 173

*t*, 174–76cognitive feedback, 175

outcome feedback, 174–75

performance feedback, 175

task properties feedback, 175–76

fertility intentions, interpreting, 461–62

FIGARCH (fractionally integrated GARCH) model, 510

*f*, 515final forecasts, based on judgment, 166–67

difficulty of, 521

long-horizon, 517–19

random walk hypothesis and, 501–6

Technical analysis and, 500

volatility forecasting and, 506–13

finite-sample predictive ability tests, 423–26

Clark/McCracken study, 425–26

Giacomini/White study, 423–24

null hypothesis and, 417

first arc sine law of probability, 502

fiscal authority, NAWM, 92

fixed dummy seasonal component, 133–34

“fixed forecasting scheme,” 442

fixed trigonometric seasonal component, 134–35

Fletcher.

*See*BFGS methodforecast combinations, 3, 355–88.

*See also*model instabilityassigning weights in, 356

data revisions/real-time data, 263

design of model universe in, 356

DFMs and, 306–7

empirical studies, 305–6

GVARs and, 305–6

misspecification biases and, 386

model instability's impact on, 357

as pooling, 301

portfolio theory

*v*., 304–5risk and, 386

VARs and, 305

forecast combinations illustration: survey and time series forecasts, 3, 358–61

design of model universe, 358–61

equal-weighted combination scheme, 359–61

individual time series models, in universe of models, 361

pseudo out-of-sample forecasts and, 364

restricted universes of models, 361

survey forecasts, in universe of models, 361

forecast densities, nonlinear models and, 84

(p. 698)
forecast error taxonomies, 277–79

data measurement errors, 278

location shifts, 278

misspecification of in-sample deterministic factors, 278

misspecification of stochastic components, 278

forecast evaluation, 3–4.

*See also*conditional predictive ability tests; forecast performance assessment; multidimensional panel data of survey forecasts; multiple forecast model evaluation; survey forecasts; unconditional predictive ability testsforecasting (macroeconomic forecasting).

*See also*election forecasting; financial time series forecasting; nonlinear time series forecasting; sales forecasting; volatility forecastingcointegration's impact on, 16–17

models

*v.*, 287weather and climate forecasting

*v.*, 561forecasting breaks, 3, 315–53.

*See also*forecasting during breaks; forecasting models; informationforecasting regime shift, 336–39

predictability and, 317

strategy outline, 319–20

forecasting during breaks, 336.

*See also*forecasting breaksforecasting during new exponential break, 345–46

forecasting during new location shift, 339–41

Monte Carlo analysis, 341–45

forecasting in presence of unanticipated location shifts (with misspecified models), 271–314.

*See also*forecast error taxonomiesdata revisions and structural breaks, 307–8

econometric models

*v*. robust forecasting models, 279–88forecast error taxonomies, 277–79

forecasting models, for forecasting breaks.

*See also*forecast combinations; forecasting breaksapproximating logistic transition function, 330–31

conventional, 328–29

formulation of, 327–31

IIS and, 333–34

modeling nonlinearity, 334–35

nonlinear functions, 329

reduction to theory-based form, 335–36

testing for nonlinearity, 331

forecast performance assessment.

*See also*forecast combinations; forecasting in presence of unanticipated location shiftsgeneral purpose loss functions for, 274–75

GFESM measure and, 276–77

MSFE matrix and, 275–77

NAWM, 107

relative accuracy in, 277

RMSFEs and, 277

forecast uncertainty

density forecasts and, 113

multidimensional panel data of survey forecasts and, 481–85

fractional cointegration, 521

fractional marginal likelihood criterion, 16

“framing” effects, 172

frost-fruit case study, 572

fruit-frost case study, 572

G7 countries data revisions (study), 254

Galton, F., 465

GARCH (generalized autoregressive conditional heteroskedastic) models

ACD model

*v*., 518–19ARFIMA models

*v*., 540–41controlling for FWER, 393

discrete-time, 543

GARCH-in-mean, 519

HYBRID GARCH structure, 549

parallel GARCH structure, 547

Realized GARCH model, 547–49

gas market.

*See also*electricity spot marketscarbon emissions markets and, 609

oil pricing and, 608

pricing, 608

VARs and, 608

Gaussian MLEs.

*See*maximum likelihood estimationsGaussian state-space models.

*See*state-space modelsGDP.

*See*long-horizon economic growth forecasting; nowcasting modelgeneralized autoregressive conditional heteroskedastic models.

*See*GARCH modelsgeneralized Bass model, 682

generalized linear models.

*See*GLMsgeneral polynomial approximation [Ch11, Section 6.3]

general unrestricted model(GUM), 335–36

generational accounting, 585

GFESM (generalized forecast error second moment) measure, 20, 21, 22t, 23, 276

forecast performance assessment and, 276–77

GLMs (generalized linear models for health care costs), 5, 627, 636–39, 642, 642

*t*, 645, 646–47, 648, 648*t*, 649, 649*t*basis approach, 636–39

global vector autoregressive (GVAR) models, 305–6

Goldfarb.

*See*BFGS method“Great Moderation,” 367

GUM (generalunr estricted model), 335–36

GVAR (global vector autoregressive) models, 305–6

Hannan-Quinn criterion.

*See*HQ criterionHARCH (heterogeneous ARCH) model, 541

Harmonized Index of Consumer Prices.

*See*HICPHealth and Retirement Study (HRS), 464

health care costs models, 5, 625–54

administrative data sets for, 627

binary models, 626

challenges for, 625

cost effectiveness analysis and, 626

count data models, 626

log transformations, 629–31

negative binomial model, 633

risk adjustment and, 626

semiparametric transformation models, 632

square root transformations, 631

usage areas for, 626

HEAVY model, 547

Herfindahl–Hirschman index (HHI), 619

heterogeneous ARCH (HARCH) model, 541

heterogeneous autoregressive model of realized variance.

*See*HAR-RVheterogeneous autoregressive models.

*See*HAR modelsheterogeneous survey forecasts.

*See*survey forecastsheuristics, 168–70

HHI (Herfindahl–Hirschman index), 619

hidden Markov regression models.

*See*Markov switching regression modelshierarchical DFMs, 54

high-frequency data.

*See also*disaggregation by time; disaggregation over variables; volatility forecastingbreak detection and, 326

dynamic properties of volatility and, 526–27

SV modele stimation and, 528

volatility forecast evaluation and, 527

volatility forecasting impacted by, 525–26

Hooker, R., 471

HRS (Health and Retirement Study), 464

HYBRID GARCH structure, 549

IEEE Monthly Business Survey, 467

“I knew it all along effect,” 171

illusion of control, 171

impulse-indicator saturation.

*See*IISIndian Ocean tsunami, 316–17

information (for predicting breaks), 317.

*See also*information setsconditions for forecasting breaks and, 316–18

disaggregated data over time and variables, 326

formal description of, 323–24

improved data at forecast origin, 327

leading indicators, 325–26

new sources for, 319

prediction markets, 326–27

separation into sets, 320

survey data, 326

information criteria.

*See also*AIC; BIC; HQ criterionestimation of static factors with, 46–47

VARs and, 15–16

instrumental variables, dynamic factors as, 50

integrated GARCH.

*See*IGARCH modelsinterpreting and combining survey forecasts.

*See*survey forecastsjudgment (in economic forecasting), 2, 163–89.

*See also*biases“black arts” and, 174

final forecasts based on, 166–67

in formulation of models, 164–65

news revisions

*v*. noise revisions issue and, 261prediction markets and, 182

in revising components of models, 166

“twilight world” and, 174

Kalman filter.

*See also*maximum likelihood estimationsMIDAS regressions

*v*., 236–38recursive lemma and, 141–42

KG polynomials.

*See*Kolmogorov-Gabor polynomialslarge-scale forecast comparisons, 78–81

leads, MIDAS with, 235–36

lemma, Kalman filter and, 141–42

likelihood evaluation.

*See*maximum likelihood estimationsLindley's statistical paradox, 113

linear Gaussian state-space models.

*See*state-space modelslinear-quadratic models, 122

linear regression health care costs models, 626, 627–32.

*See also*transformed health care costs modelscost regression models, 627–29

tests for, 628–29

transformed costs models, 629–32

linear time series models.

*See also*forecast combinations illustration: survey and time series forecasts; nonlinear time series modelsstate-space form and, 138

linear time series/nonlinear time series forecasts comparison, 61, 76–77

forecasting with same model for each forecast horizon, 80–81

forecasting with separate model for each forecast horizon, 78–80

large-scale forecast comparisons, 78–81

local DGPs.

*See*LDGPslocal linear trend component, 132

location shifts.

*See also*forecasting during breaks; forecasting in presence of unanticipated location shifts; structural breaksdata revisions and, 307–8

data vintages and, 307–8

DGPs and, 280–81

forecast error taxonomy and, 278

improving robustness to, 283

VEqCMs impacted by, 282–83

logistic STAR.

*See*LSTAR modelslogistic STR.

*See*LSTR modelslog-linearized DSGEs.

*See*DSGEs, log-linearizedlog transformations, cost regression health care models, 629–31

long-horizon economic growth forecasting (demographically based), 5, 585–605.

*See also*demographyadvantages of, 586

benefits of, 602

economic growth theory and empirics, 587–90

GDP development and, 585–602

uncertainty measures of, 603

long-horizon financial time series forecasting, 517–19

long-memory stochastic volatility (LMSV) model, 540

LSTAR (logistic STAR) models.

*See also*forecast combinations illustration: survey and time series forecastsneuralnetw orks and, 515–17

“lucky” model, 392

macroeconomic forecasting.

*See*forecastingMaori/Captain Cook example, 321

Markov switching (MS) regression models, 45, 61, 65–66, 383

ARFIMA and, 543

GNP and, 77

as nonlinear time series models, 65–66

maximum likelihood estimations (MLEs)

DSGE estimation, 52

nowcasting model, 201–2

state-space models, 142–43

mean absolute percentage errors.

*See*MAPEmean error.

*See*MEmean forecasts, convex loss functions and, 465–66

mean prediction error.

*See*MPEmean squared error.

*See*MSEmedian forecasts, unimodal loss functions and, 466

Meese-Rogoff puzzle, 446

MEM (multiplicative error model), 547

Michigan Monthly Survey, 464

*Microeconometrics Using Stata*data set, 644

MIDAS (mixed-data sampling) regressions, 227–35, 241

ADL-MIDAS, 231–35

direct/iterated approaches

*v*., 226HAR-RV and, 239–40

HYBRID GARCH structure and, 549

Kalman filter

*v*., 236–38with leads, 235–36

Matlab Toolbox for, 241

MIDAS-NIC, 240–41

MIDAS-RV, 240

missing data.

*See also*EM algorithm; Kalman filtermisspecification

for DSGEs, 124

forecast combinations and, 386

of in-sample deterministic factors, 278

for NAWM, 124

of stochastic components, 278

misspecified models.

*See also*forecast error taxonomies; forecasting in presence of unanticipated location shiftsforecast error taxonomies and, 277

structural breaks and, 302–5

mixed-data sampling.

*See*MIDAS regressionsmixed-frequency data, 225–45.

*See also*MIDAS regressionscomputer technology innovations and, 225

forecasting with, 225–27

MLEs.

*See*maximum likelihood estimationsmodel-based volatility forecasting, 529, 543–50

realized measures and, 528

reduced-form volatility forecasting

*v*., 527–28model instability, 3, 357–58, 375–82.

*See also*structural breaksforecast combinations impacted by, 357

models.

*See also*DSGEs; dynamic factor models; forecasting models; health care costs models; judgment; nonlinear time series models; unobserved components time series models; VARs;*specific models*advantages of, 163–64

data issues and, 1

forecasting

*v.*, 287judgment's role in, 164–66

limitations, 163–64

“lucky,” 392

susceptibility of, to unanticipated structural breaks, 272

model selection [Ch11, section 6] also see Autometrics

Monte Carlo analysis.

*See also*forecast combinations experiment: in presence of breaks; Markov chain Monte Carlo methodsforecasting during breaks, 341–45

health care cost models comparison, 643

unconditional predictive ability experiments, 429–38

Monthly Business Survey, IEEE, 467

MS.

*See*Markov switching regression modelsMSFE (mean squared forecast error) matrix

forecast performance assessment and, 275–77

limitations of, 275–77

multidimensional panel data of survey forecasts, 4, 473–95

anticipated changes measures, 478–81

rationality tests, 485–92

volatility measures, 478–81

multinominal health care costs models, 626

multiple cointegrating vectors, 521

multiple cycles component, 137

multiplicative error model (MEM), 547

multivariate factor-based models (AR_FAC), 361.

*See also*forecast combinations illustration: survey and time series forecastsmultivariate time series models, 148–49.

*See also*unobserved components time series modelschallenge of, 148–49

estimation methodology, 152

forecasting methodology, 152

multivariate trend model, 149–50

state-space representation, 152

multivariate trend model, 149–50

Muth's unbiasedness condition, 474–75

National Institute of Statistics and Economic Studies (INSEE) Monthly Business Survey, 467

National Longitudinal Survey of Youth, 464

NAWM (New Area-Wide Model), 91–101

bird's eye view of, 92–93

empirical implementation of, 96–101

fiscal authority, 92

forecast accuracy evaluation, 107

key equations in, 93–96

misspecification matters for, 124

performance/structure relationship, 117–22

nearest neighbor forecast method, 76

negative binomial model, 633

neural networks.

*See also*forecast combinations illustration: survey and time series forecastsautoregressive, 79

forecasting with, 330

regime switching models and, 515–17

RMSFE, 72

‘single hidden-layer feedforward’ model, 70

as universal approximators, 70–72

New Area-Wide Model.

*See*NAWMNIC.

*See*news impact curvenoise revisions, news revisions

*v.*, 260–61nonlinear DFMs, 45–46

nonlinear functions, 329

(p. 705)
nonlinearity.

*See also*forecasting models; volatility forecastingin reduced-form volatility forecasting, 543

testing for, 331

nonlinear/non-Gaussian state-space models, 45–46

nonlinear threshold autoregressive model, 77

nonlinear time series/linear time series forecasts comparison, 61, 76–77

forecasting with same model for each forecast horizon, 80–81

forecasting with separate model for each forecast horizon, 78–80

large-scale forecast comparisons, 78–81

nonlinear time series models, 61–87.

*See also*forecast combinations illustration: survey and time series forecasts; Markov switching regression models; neural networks; STAR models; STR models; switching regression models; TAR modelsDFMs

*v.*, 2forecast densities and, 84

random coefficient models, 67–68

VARs

*v.*, 2Nordhaus's efficiency condition, 474–75

Nord Poolel ectricity market.

*See*electricity daily spot pricesnowcasting, 193–223.

*See also*nowcasting modelautomatic statistical procedure

*v.*, 195forecast failure, 163

future research on, 212–13

high-frequency predictors and, 203

importance, in forecasting literature, 195

new developments, 212–13

related literature and, 202–4

single equation approaches, 203

structural breaks and, 327

surveys and, 194

VARs and, 212

nowcasting model (euro-area GDP), 213

benchmark model, 205

empirical results, 204–12

forecast updates, 207–9

key feature, 202–3

limitations, 212–13

maximum likelihood estimation, 201–2

monthly factor model, 199–200

quarterly variables model, 200–201

reasons for GDP emphasis, 194

null hypothesis.

*See also*FDR; finite-sample predictive ability tests; population-level predictive ability testsalternative hypothesis and, 391

data snooping and, 391–92

defined, 411

false discovery and, 392

forms of, 417

objective probability of personal outcomes, point predictions and, 460–61

oil market.

*See also*electricity spot marketscarbon emissions markets and, 609

cointegration analyses and, 608

energy commodities bundle and, 607–8

models/conceptual frameworks for, 607–8

pricing, VARs and, 608

volatility of, 620

OLG (overlapping generations) model, 589

OPEC (Organization of Petroleum Exporting Countries), 607

Organization of Petroleum Exporting Countries (OPEC), 607

Ornstein–Uhlenbeck process, 526

outcome feedback, 174–75

overlapping generations (OLG) model, 589

overparameterization problem, 9–10

ox weighing, 465

panel-data models, 337

panel-data sets, 467, 473.

*See also*multidimensional panel data of survey forecasts; SPF; survey forecastsBank of England's Survey of External Forecasters, 467

Center for Economic Studies ifo Institute World Economic Survey, 467

National Institute of Statistics and Economic Studies (INSEE) Monthly Business Survey, 467

paradox, Lindley's, 113

parallel GARCH structure, 547

PCCA.

*See*partial canonical correlation analysisperformance assessment.

*See*forecast performance assessmentperformance feedback, 175

PIC.

*See*posterior information criterionplanting/fallowing case study, 572

point predictions.

*See also*survey forecastsof binary outcomes, 459–62

cross-forecaster dispersion, interpreting, 462–63

of fertility intentions, 461–62

objective probability of personal outcomes and, 460–61

of real-valued events, 462–63

subjective probabilities and, 459–60

policy

differenced VEqCM and, 287

DSGEs and, 2

DSGE-VARs and, 124

forecasting

*v.*, 287health care costs models and, 626

models

*v.*, 287quantitative models and, 164

stochastic breaks and, 278

pooling.

*See*forecast combinationspopulations, stable, 589–90

portfolio theory, forecast combinations

*v.*, 304–5posterior information criterion (PIC), 16

power markets.

*See*electricity spot marketspredictive ability testing.

*See*conditional predictive ability tests; unconditional predictive ability testsprequential approach, 113

probability forecasting.

*See also*multidimensional panel data of survey forecasts; survey forecasts; weather and climate forecastselection forecasting and, 657

learning through feedback and, 174

prediction markets and, 327

in survey forecasts, 463–64

weather/climate forecasting and, 562–64

(p. 707)
pseudo out-of-sample forecasting exercises

defined, 416

DFMs and, 49

forecast combinations illustration and, 364

vintage data comparisons and, 307

“publicity hypothesis,” 167

*p*-value, 406

quantitative models.

*See*modelsragged edge.

*See*jagged edge problemrandom coefficient models, 67–68

random walk benchmark.

*See*NAWM benchmarksrandom walk hypothesis, 501–6

random walk plus smooth cycle model, 140–41

random walk seasonal component, 135–36

RC (reality check) procedure.

*See also*SPA testdefined, 392–93

SPA test

*v.*, 398stepM approach

*v.*, 399–400reality check.

*See*RC procedureRealized GARCH model, 547–49

realized measures

benefits of, 526

defined, 525

model-based volatility forecasting and, 528

reduced-form volatility forecasting and, 527–28

of volatility, 531–32

“realized volatility,” 507

real-time data.

*See also*data revisionsdata revisions

*v.*, 257–59real-time data analysis bibliography (online), 254

real-time datasets (online), 252

unconditional predictive ability tests and, 428–29

recursive lemma, Kalman filter and, 141–42

recursive regression strategies, 520

reduced-form models, 90

reduced-form volatility forecasting, 527–29, 537–43, 549–50

distributed lag models, 537–40

model-based volatility forecasting

*v.*, 527–28nonlinearity in, 543

realized measures and, 527–28

separating continuous and jump components, 542–43

regression coefficients, 143–44

regression component.

*See*irregular componentrelative accuracy, in forecast performance assessment, 277

relative mean absolute revision (RMAR), 253

representativeness heuristic, 168–69

“reputational cheap talk” hypothesis, 167

residential in vestment.

*See*data revisions examplerevisions.

*See*data revisionsrisk, forecast combinations and, 386

risk adjustment, health care costs models and, 626

RMAR (relative mean absolute revision), 253

RMSEs.

*See*root mean squared errorsRMSFEs.

*See*root mean squared forecast errorsrobust forecasting models, econometric models

*v.*, 279–88
(p. 708)
rolling window estimates, 18, 26, 28, 29

*t*, 305, 394n2, 395, 424, 447.*See also*expanding window estimatesroot mean squared forecast errors (RMSFEs)

forecast performance assessment and, 277

neural networks and, 72

roulette wheel, 501

sampling the future method, 90.

*See also*NAWMalternative forecasting models, 103–7

estimating predictive distribution of log-linearized DSGE model, 101–3

scientific election forecasting.

*See*election forecastingseasonal component, 129

fixed dummy, 133–34

fixed trigonometric, 134–35

random walk, 135–36

specifications, 133–36

time-varying dummy, 134

time-varying trigonometric, 135

second-stage regressions, dynamic factors and, 48–50

SEE (Survey of Economic Expectations), 464

self-exciting threshold autoregressive models.

*See*TAR modelssemiparametric approaches.

*See*discrete conditional density estimator; finite mixture health care costs modelssemiparametric transformation models, 632

sequential testing bias problem, 392

Shannon.

*See*BFGS methodshocks.

*See also*energy commodities; multidimensional panel data of survey forecastsdiscrete, 478–81

ʼsingle hidden-layer feedforward’ model, 70

“single source of error” models, 12

SM-AR (switching-mean autoregressive) model, 71

smooth transition autoregressive models.

*See*STAR modelssmooth transition regression models.

*See*STR modelsspecial interest areas, 5.

*See also*election forecasting; energy commodities; health care costs models; long-horizon economic growth forecasting; sales forecasting; weather and climate forecastsSPF (Survey of Professional Forecasters)

square root transformations, cost regression health care models, 631

SR models.

*See*switching regression modelsstable populations, 589–90

STAR (smooth transition autoregressive) models, 64.

*See also*LSTAR modelsneural networks

*v.*, 516–17vector, 65

state-space models, 138

diagnostic checking, 144–46

with dynamic factors, 43–44

likelihood evaluation, 142

linear time series models and, 138

log-linearized DSGE, 90

MLE of, 142–43

multivariate version (unobserved components time series models), 152

nonlinear/non-Gaussian, 45–46

parameter estimation, 142–44

with static factors, 43

time-domain maximum likelihood (via Kalman filter) and, 38–39

static factors, 38.

*See also*dynamic factorsdetermining number of, 46–47

estimation of dynamic factor innovations from, 42

exact static factor model, 41

state-space model with, 43

statistical models.

*See*election forecasting statistical models; sales forecastingstepwise multiple testing approach (stepM), 399–400

stochastic models, of electricity spot prices, 615–17

stochastic volatility (SV) models, 508–9

estimation, high-frequency data and, 528

IGARCH(1,1) models

*v.*, 508–9stop-break model, 336

straw polls, 657

strong form reduced-rank structure, 15

structural breaks (unanticipated structural breaks).

*See also*forecasting breaks; forecasting during breaks; forecasting in presence of unanticipated location shifts; location shifts; model instabilityas “absorbing barrier,” 328

causes of, 272

DFMs with, 52–53

DGPs and, 272

disaggregation by variable and, 291

misspecified models and, 302–5

models' susceptibility to, 272

nowcasting and, 327

testing, 447

structural time series models, 131

structural VARs.

*See*SVARssubjective probabilities, point predictions and, 459–60

superior predictive ability test.

*See*SPA testsurvey forecasts (interpreting and combining), 4, 457–72.

*See also*forecast combinations illustration: survey and time series forecasts; multidimensional panel data of survey forecastsmean forecasts and convex loss functions, 465–66

nowcasting and, 194

point predictions of binary outcomes, 459–62

point predictions of real-valued events, 462–63

probabilistic forecasting, 463–64

Survey of Economic Expectations (SEE), 464

Survey of Professional Forecasters.

*See*SPFSV.

*See*stated value studies; stochastic volatility modelsSweden economic growth studies, 597–98, 597

*f*, 600–601, 601*f*.*See also*long-horizon economic growth forecastingswitching-mean autoregressive (SM-AR) model, 71

task properties feedback, 175–76

taxonomy, for forecast errors.

*See*forecast error taxonomiestests.

*See*conditional predictive ability tests; health care costs models; unconditional predictive ability tests;*specific tests*three-dimensional panel data of survey forecasts.

*See*multidimensional panel data of survey forecaststime disaggregation.

*See*disaggregation by timetime-domain estimation, of dynamic factors, 37–44

time-varying dummy seasonal component, 134

time-varying parameters, DFMs with, 52–53

time-varying trigonometric cycle component, 136–37

time-varying trigonometric seasonal component, 135

Tracy–Widom law, 47

transformed health care costs models, 629–32

log transformations, 629–31

semiparametric transformation models, 632

square root transformations, 631

trend component, 129.

*See also*smooth trend componentlocal linear, 132

multivariate, 149–50

specifications, 131–33

with stationary drift, 132

trigonometric components

fixed trigonometric seasonal component, 134–35

smooth time-varying trigonometric cycle component, 137

time-varying trigonometric cycle component, 136–37

time-varying trigonometric seasonal component, 135

tsunamis, 316–17

“twilight world,” 174

2004 Indian Ocean tsunami, 316–17

UC-RV (unobserved components-realized volatility) model, 541

UK election forecasting models, 662–63

unanticipated structural breaks.

*See*structural breaksunbiasedness condition, Muth's, 474–75

uncertainty.

*See also*forecast uncertainty; parameter estimation uncertaintyuncertainty measures of demographically based long-horizon growth forecasting, 603

(p. 711)
unconditional predictive ability tests, 4, 415–40

data revisions and, 428–29

existing extensions of research, 427–28

Monte Carlo experiments, 429–38

other dimensions, 427–29

real-time data and, 428–29

topics for future research, 429

unimodal loss functions, median forecasts and, 466

univariate model (AR), 361, 365

*t*, 366*t*.*See also*forecast combinations illustration: survey and time series forecastsunmodeled breaks.

*See*location shifts; structural breaksunobserved components-realized volatility (UC-RV) model, 541

unobserved components time series models, 129–62.

*See also*cycle component; irregular component; seasonal component; trend componentautoregressive representation of, 148

extending, to non-Gaussian class, 159

forecasting, 146–48

interest variable forecasting and, 130

observation weights of forecast function, 147–48

unrestricted VARs.

*See*UVARsvaluation of weather and climate forecasts.

*See*weather and climate forecastsVARFIMA (vector fractionally integrated ARMA) model, 541

variable of interest forecasting, 130

variables disaggregation.

*See*disaggregation over variablesVARs (vector autoregressive models), 1–2, 9–34.

*See also*BVARs; NAWM benchmarks; SVARsforecast combinations and, 305

gas pricing and, 608

GVARs, 305–6

information criteria and, 15–16

Minnesota prior and, 103–4

nonlinear time series models

*v.*, 2nowcasting and, 212

oil price dynamics and, 608

overparameterization problem in, 9–10

sequential testing and, 15–16

VEC (vector error-correction), 10

VECMs (vector error-correction models)

Bayesian, 25

Bayesian reduced-rank, 18–19

BVARs and, 19

cobreaking in, 17

vector autoregressive models.

*See*VARsvector equilibrium correction models.

*See*VEqCMsvector error-correction.

*See*VECvector error-correction models.

*See*VECMsvector fractionally integrated ARMA (VARFIMA) model, 541

vector STAR models, 65

vector switching regression model, 63

vector threshold autoregressive model, 63

VEqCMs (vector equilibrium correction models), 280

DDD

*v.*, 283–87in-sample forecast model, 281–82

LDGP nonconstancy and, 284–87

location shift's impact on, 282–83

verification, weather and climate forecasts, 562–63

volatility

assessing persistence of, 533–35

dynamic properties, high-frequency data and, 526–27

measuring, 531

multidimensional panel data of forecasts and, 478–81

of oil market, 620

realized measures of, 531–32

“realized volatility,” 507

statistical properties of, 523–35

volatility forecasting

financial time series forecasting and, 506–13

GARCH models and, 507–13

long memory models and, 513–15

methods for, 507

“realized volatility” and, 507

survey of, 507

SV models and, 508–9

volatility forecasting, using high-frequency data, 4, 525–56

data cleaning, 532

distributed lag models, 537–40

future research, 550

GARCH models and, 543–49

notational framework, 529–30

volcanic eruptions, 318

*vox populi*, 465

weak form reduced-rank structure, 14

weather and climate forecasts (economic value assessment), 5, 559–83

background on methods, 560–64

case studies, 570–73

climate

*v.*weather forecasts, 561economic forecasting

*v.*, 561“good,” 562

probability, 562–64

prototypical decision-making models, 567–70

reliability and, 562–63

revealed preference methods, 573

verification, 562–63

weather

*v.*climate forecasts, 561weights

assigning, in forecast combinations, 356

equal-weighted combination scheme (forecast combinations illustration), 359–61

equal-weighted forecast

*v.*best model forecast (model instability illustration), 381–82, 381*f*, 386–87observation weights of forecast function (unobserved components time series models), 147–48

ox weighing and, 465

Wharton Research Data Services (WRDS) system, 534

white chaos, 517

“wishful expectations hypothesis,” 168

Wold forecast error, 197

Wold representation, 12

World Economic Survey, Center for Economic Studies ifo Institute, 467

WRDS (Wharton Research Data Services) system, 534

YADA program, 96