Economic Modelling 27 (2010) 14531462

Contents lists available at ScienceDirect

Economic Modelling
j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / e c m o d

The use of consumer and business surveys in forecasting

Lawrence R. Klein, Sleyman zmucur
University of Pennsylvania, Department of Economics, 3718 Locust Walk, Philadelphia, PA 19104, USA

a r t i c l e
JEL classication: C22 C42 C52 C53 E37 Keywords: Surveys Forecasting Evaluation European Union

i n f o

a b s t r a c t
Surveys improve forecasting performance by adding explanatory power to a model which is based on only past values of manufacturing growth. The issue addressed in this paper is whether surveys of production expectations, when added to equations that contain lagged values of a headline index pertaining to the real economy, improve forecasting performance. If so, it may be better for researchers to use not just the headline index, but production expectations or the Economic Sentiment Indicator if they wish to better predict manufacturing growth. 2010 Elsevier B.V. All rights reserved.

1. Introduction Global recession has proved that no country or group of countries, including the Euro area1 and the European Union,2 is immune to adverse developments in other countries. The drop in industrial production in the European Union (EU27) was unprecedented in recent history (Fig. 1). After ve three-year growth cycles during the post-1990 era, a freefall in production started in early 2008. The index of manufacturing, which was 80.47 (2005 = 100 in January 1990), reached its peak of 113.34 in February 2008, and declined to 90.27 in July 2009.3 The costs of these developments in production and employment are immense and bring up the question of the predictability of such drops in economic activity and the usefulness of those forecasts for decision makers. Although nancial data are available in real time, most of the variables related to real side of the economy are available only monthly or even quarterly (in case of gross domestic product). On the other

hand, most economic analysis involves real side of the economy. This situation naturally causes delays in any realistic empirical work. The lag in the publication of data pertaining to the real economy compounds the problem.4 For example, June industrial production in the European Union was released on August 12th (about six weeks after the end of the month). On the other hand, consumer and business surveys, which have been conducted in member countries of the European Union, are released at the end of every month.5 For example, August survey results were released on August 28th. There is about a two-month lead in the release of survey data compared with the release of industrial-production gures. The timeliness in data release is very important, especially at times of great uncertainty. It increases the signicance of business and consumer survey results in forecasting even further (Kauppi et al., 1996; Klein and zmucur, 2002a,b). This situation necessitates the use of survey results in forecasting, the primary rationale

Corresponding author. Tel.: + 1 215 898 6765; fax: + 1 215 898 4477. E-mail address: ozmucur@econ.upenn.edu (S. zmucur).
1 There are 16 member states in the Euro area (EA): Belgium, Germany, Ireland, Greece, Spain, France, Italy, Cyprus, Luxembourg, Malta, the Netherlands, Austria, Portugal, Slovenia, Slovakia, and Finland. 2 There are 27 member states in the European Union (EU27): Belgium, Bulgaria, the Czech Republic, Denmark, Germany, Estonia, Ireland, Greece, Spain, France, Italy, Cyprus, Latvia, Lithuania, Luxembourg, Hungary, Malta, the Netherlands, Austria, Poland, Portugal, Romania, Slovenia, Slovakia, Finland, Sweden, and the United Kingdom. 3 The arithmetic average growth rate for the entire period is, 1.0%, with a maximum rate of 8.8% (December 1994) and a minimum rate of negative 19.2% (February 2009).

4 Since many countries are involved, lags in economic data releases are longer in the European Union than in the United States. For example, July industrial production was released on August 14, 2009 in the US while June industrial production in the EU was released on August 12, 2009. 5 There are ve surveys of consumer and business (with varying numbers of questions). They pertain to: Industry, Construction, Retail, Services, and Consumers. Results of these surveys are used to construct Condence Indicators, which are then used to produce an Economic Sentiment Index (ESI) and a Business Climate Indicator (BCI). For a very detailed history and account of these surveys and workshop papers, see European Commission Directorate-General for Economic Financial Affairs (2006, 2007, 2009b). See, also Hansson et al. (2005), Kauppi et al. (1996), and Lemmens et al. (2005).

0264-9993/$ see front matter 2010 Elsevier B.V. All rights reserved. doi:10.1016/j.econmod.2010.07.005

1454

L.R. Klein, S. zmucur / Economic Modelling 27 (2010) 14531462

Fig. 1. European Union (EU27) manufacturing index (2005 = 100) and its annual growth (%).

for this paper. Specically, we try to answer the following questions: 1. Do survey results contribute to accuracy in forecasting the variable of interest? 2. Although researchers tend to concentrate on the headline index, is there a better predictor among responses than the headline gure? Our empirical results indicate that answers to both of these questions are positive. The paper is in ve sections. The model is presented in the next section. The third section is devoted to empirical results that arise from estimating the model. The fourth section contains an evaluation of forecasts. The nal section contains the major conclusions drawn from this research. 2. The model Surveys are said to be useful if they increase the explanatory power of the forecasting equation. A forecasting equation may be in the form of an autoregressive-integrated-moving-average model (Deitz and Steindel, 2005), and/or include a principal component (Klein and zmucur, 2002a,b, 2004), or VAR, TAR, SETAR, Markov Switching (Claveria et al., 2007). Here, the simplest form proposed is (AR (12)), a specication that allows a researcher to obtain a forecast without much effort.6 The model to be used is in the following form (icountry, and ttime): Yit = ij Yitj + i Sit + it ; i = 1; 2; 27; EA; EU; j = 1; 2; 12; t = 1991M012009M06 where, Y is the year-on-year growth7 of the economic indicator of interest (industrial production, construction output, retail sales, services output, and consumer expenditures8), and S is the indicator
6 The model that is used in testing the signicance of survey results is very similar to the one used by Deitz and Steindel (2005). 7 Year-on-year growth is called annual growth by Eurostat. 8 In contrast to the U.S., data on consumer expenditures are not available monthly in the European Union.

from business and consumer surveys [condence index (industrial production, construction output, retail sales, services output, or consumers) or balance on a survey question], and is the stochastic disturbance term, which is assumed to be iid. In addition to the benchmark model (AR (12)), ten alternative models are estimated. These models are based on indices constructed from answers to individual questions in surveys of consumer and business condence. There are seven questions in the monthly industry survey. These questions are presented in the succeeding data. In addition to these seven questions, the Industrial Condence Index, Economic Sentiment Index, and a proxy for the Business Conditions Index, are used in some of the models shown. Alternative models
Model Benchmark model, AR (12), Eq. (1) without the variable S Models 1 through 10 include one additional variable, in addition to AR (12) 1 Q1. How has production developed over the past 3 months (increased, unchanged, decreased)? 2 Q2. Current overall order books (above normal, normal for the season, below normal) 3 Q3. Current export order books (above normal, normal for the season, below normal) 4 Q4. Current stock of nished products (above normal, normal for the season, below normal) 5 Q5. How do you expect production to develop over the next 3 months (increase, unchanged, decrease)? 6 Q6. How do you expect selling prices to change over the next 3 months (increase, unchanged, decrease)? 7 Q7. How do you expect rm's total employment to change over the next 3 months (increase, unchanged, decrease)? 8 Business Conditions Index,a (Q1 + Q2 + Q3 + Q5 Q4) / 5 9 Industrial Condence Indicator, (Q2 + Q5 Q4) / 3 10 Economic Sentiment Indicator, ESI (weighted average of 15 components in industry (3), services (3),consumers (4), construction (2), and retail trade (3) condence indicators)
a This is a variable constructed by us, and not to be confused with the Business Conditions Index released regularly by the European Commission for the euro area, though the same variables are used.

L.R. Klein, S. zmucur / Economic Modelling 27 (2010) 14531462 Table 1 Adjusted determination coefcients in alternative models. 0. AR (12) Austria Belgium Bulgaria Cyprus Czech Republic Denmark Estonia Finland France Germany Greece Hungary Ireland Italy Latvia Lithuania Luxembourg Poland Portugal Romania Slovakia Slovenia Spain Sweden The Netherlands UK Euro area (EA16) European Union (EU27) 0.794 0.814 0.774 0.432 0.815 0.558 0.911 0.805 0.882 0.844 0.623 0.841 0.482 0.866 0.890 0.455 0.600 0.760 0.613 0.869 0.783 0.798 0.803 0.910 0.626 0.842 0.927 0.935 1. Production 2. Orders 3. Export orders 0.799 0.833 0.772 0.447 0.819 0.589 0.917 0.821 0.886 0.869 0.627 0.848 0.475 0.878 0.898 0.468 0.574 0.768 0.656 0.834 0.794 0.837 0.830 0.909 0.644 0.858 0.931 0.937 4. Stocks 5. Production expectations 0.822 0.840 0.772 0.457 0.836 0.600 0.917 0.829 0.898 0.887 0.689 0.862 0.469 0.875 0.913 0.511 0.584 0.769 0.667 0.869 0.804 0.834 0.842 0.912 0.631 0.876 0.947 0.955 6. Price expectations 0.796 0.831 0.774 0.447 0.821 0.589 0.911 0.811 0.884 0.864 0.623 0.854 0.462 0.867 0.889 0.459 0.570 0.762 0.651 0.833 0.785 0.807 0.831 0.910 0.651 0.858 0.927 0.934 7. Employment expectations 0.800 0.838 0.775 0.445 0.825 0.611 0.924 0.818 0.886 0.877 0.646 0.841 0.473 0.867 0.903 0.470 0.566 0.761 0.647 0.869 0.799 0.841 0.830 0.910 0.651 0.860 0.931 0.937 8. Business Conditions Index 0.805 0.837 0.771 0.485 0.831 0.598 0.929 0.834 0.888 0.885 0.663 0.852 0.483 0.878 0.905 0.489 0.583 0.783 0.657 0.835 0.807 0.841 0.833 0.903 0.645 0.863 0.936 0.942 9. Industrial Condence Indicator 0.806 0.835 0.771 0.499 0.828 0.602 0.926 0.832 0.888 0.884 0.673 0.852 0.481 0.875 0.903 0.495 0.584 0.777 0.655 0.869 0.800 0.837 0.833 0.903 0.640 0.868 0.937 0.943

1455

10. Economic Sentiment Indicator 0.829 0.838 0.777 0.493 0.837 0.592 0.926 0.833 0.886 0.876 0.700 0.849 0.498 0.873 0.898 0.509 0.589 0.763 0.657 0.869 0.799 0.847 0.846 0.910 0.639 0.874 0.943 0.950

0.807 0.838 0.772 0.465 0.841 0.577 0.923 0.826 0.888 0.879 0.655 0.844 0.483 0.879 0.906 0.482 0.578 0.804 0.636 0.834 0.795 0.830 0.838 0.911 0.639 0.860 0.941 0.944

0.801 0.832 0.771 0.439 0.819 0.599 0.924 0.817 0.885 0.877 0.654 0.842 0.482 0.875 0.896 0.466 0.574 0.768 0.652 0.868 0.793 0.832 0.831 0.909 0.643 0.860 0.932 0.938

0.793 0.825 0.774 0.506 0.813 0.557 0.914 0.819 0.883 0.872 0.631 0.843 0.471 0.866 0.889 0.472 0.579 0.759 0.611 0.872 0.783 0.799 0.830 0.903 0.638 0.862 0.930 0.936

3. Empirical results The model is estimated for the January 1991June 2009 period, provided data are available.9 Data are obtained from the European Commission Directorate-General for Economic Financial Affairs (2009a), and Eurostat (2009). Survey results add to the explanatory power of an autoregressive model. In the 28 cases (26 member countries,10 euro area and European Union) considered here, in only a single country (Luxembourg) does the corrected (or adjusted) determination coefcient not increase with the addition of a survey indicator (Table 1).11 Furthermore, it is not the headline index, but the production-expectations indicator, (question 5 in industrial surveys) that is the best predictor among those considered here. The model with the production-expectations indicator has the highest adjusted determination coefcient in 11 out of 28 cases (Table 1 and Table 2). This outcome should not be a surprise, given a cursory look at Fig. 2. What is even more interesting is that the Industrial Condence Indicator is not the most-important variable in the forecast pertaining to any member state. On the other hand, the Economic Sentiment Indicator (ESI), which is the most comprehensive index, is the most-important variable in the

forecasts pertaining to six countries. Price expectations and ESI are the most-important variables in fteen cases (Table 2). The results do not differ very much if one uses another criterion rather than the determination coefcient. Similar results are obtained based on the standard error of the regression,12 the Akaike information criterion, and the Schwartz criterion (Table 3). The model with production expectations has the most predictive power in ten or eleven cases, and has the greatest or secondgreatest explanatory power in fteen cases. 4. Forecast evaluation We consider the signicance of the survey variable as the rst step in forecast comparisons. The most important test of a model is its ability to forecast ex ante. In what follows, ex ante forecasts for one, two, and three periods are compared. The rst two periods do not require forecasting of survey results because they are already available for the month. Models are rst estimated for the 1991M01 (or the rst period that data are available) and 2006M12 periods. Forecasts are obtained for the months of 2007M01, 2007M02, and 2007M03.13

Three variations of the model were also estimated: 1. Indicator (S) with one lag because some variables are related to expectations for future months, 2. Indicator S for the European Union if countries are regarded as parts of a single market (similar to states in the U.S.), 3. Sum of S for the country and the S for the European Union if the results pertaining to both the member state and EU are signicant. 10 Since there are no industrial-production data available for Malta, that country is not included in our analysis. 11 The determination coefcient of 0.60 is one of the lowest among all 28 cases.

Deitz and Steindel (2005) use the standard error of the regression. This procedure probably gives an unfair advantage to the model with a survey, only for the 3-period ahead forecast because that gure will be available only at the end of the month. For example, in the beginning of August, only May manufacturing production data are available. Forecasts for June, July and August have to be obtained. Although June and July surveys results are available in the beginning of August, for August gures one has to wait until the end of August. Here, August surveys will be used if they are available. This procedure gives an advantage to the model with a survey in a three-period-ahead forecast situation.
13

12

1456

L.R. Klein, S. zmucur / Economic Modelling 27 (2010) 14531462

Table 2 Rank of models based on adjusted determination coefcients. 0. AR (12) Austria Belgium Bulgaria Cyprus Czech Republic Denmark Estonia Finland France Germany Greece Hungary Ireland Italy Latvia Lithuania Luxembourg Netherlands Poland Portugal Romania Slovakia Slovenia Spain Sweden UK Euro area (EA16) European Union (EU27) Rank 1 2 3 4 5 6 7 8 9 10 Rank 1 1 or 2 1, 2 or 3 10 11 4 11 10 10 10 11 11 11 11 10 4 10 9 11 1 11 10 10 6 10 11 11 4 11 10 10 1. Production 2. Orders 3. Export orders 8 7 8 8 9 8 8 6 5 9 9 6 7 2 6 8 8 4 6 4 9 7 4 10 7 9 8 8 4. Stocks 5. Production expectations 2 1 6 6 3 3 7 4 1 1 2 1 10 4 1 1 4 10 4 1 4 2 6 2 1 1 1 1 6. Price expectations 9 9 3 7 7 7 11 10 9 10 10 2 11 9 11 10 10 2 8 7 11 9 9 7 5 10 11 11 7. Employment expectations 7 3 2 9 6 1 5 8 7 6 7 11 8 8 5 7 11 1 9 8 2 5 2 9 6 6 7 7 8. Business Conditions Index 5 5 11 4 4 5 1 1 2 2 4 3 3 3 3 4 5 3 2 2 8 1 3 5 9 4 5 5 9. Industrial Condence Indicator 4 6 10 2 5 2 3 3 3 3 3 4 6 6 4 3 3 6 3 5 5 3 5 4 10 3 4 4 10. Economic Sentiment Indicator 1 2 1 3 2 6 2 2 6 7 1 5 1 7 7 2 2 8 7 3 3 4 1 1 3 2 2 2

3 4 7 5 1 9 6 5 4 4 5 7 2 1 2 5 7 7 1 9 10 6 8 3 2 8 3 3

6 8 9 10 8 4 4 9 8 5 6 9 5 5 8 9 9 5 5 6 7 8 7 6 8 7 6 6

11 10 5 1 11 11 9 7 10 8 8 8 9 11 10 6 6 9 11 11 1 11 10 8 11 5 9 9

1 0 0 3 0 1 0 0 1 11 1 1 1

3 3 4 3 4 2 4 2 2 1 3 6 10

0 0 0 2 5 6 3 6 5 1 0 0 0

0 1 0 3 1 4 4 9 5 1 0 1 1

2 0 0 0 2 2 1 4 5 4 2 2 2

11 4 2 5 0 3 1 0 0 2 11 15 17

0 2 1 0 1 0 5 1 6 6 0 2 3

2 3 1 0 3 4 6 4 3 0 2 5 6

3 4 6 5 7 0 0 1 1 0 3 7 13

0 2 10 6 4 4 0 0 0 2 0 2 12

6 9 4 1 1 2 4 1 0 0 6 15 19

Then, the model is estimated with the ending period of 2007M01, and forecasts are obtained for the months of 2007M02, 2007M03, and 2007M04. The model with production expectations fares very well in these comparisons (Table 4).14 It has the lowest root mean square error in fteen out of 28 cases (Table 5) for one-period and two-period-ahead forecasts, and in thirteen cases in threeperiod-ahead forecasts. The model with the Economic Sentiment Index is the second-best among those estimated. The model with the Industrial Condence Index does not perform well, not even as well as the benchmark model (Tables 4 and 5, Figs. 3 and 4). Correlations between actual and ex ante forecasts reveal similar results. The model with production expectations yields the best results fteen out of 28 one-period-ahead forecasts (Tables 6 and 7). For Bulgaria and Ireland15 the benchmark model provides the best forecast.

In summary, surveys contribute to forecasting power, especially that of the production-expectations variable. Are these contributions signicant? Our experiments based on DieboldMariano statistics indicate that the model with production-expectations variable has lowest errors in ten out of the 28 cases examined relative to the results obtained with the benchmark model. However, these differences do not seem to be statistically signicant at the ninety-ve-percent condence level (Tables 7 and 8). More work has to be done on the study of the signicance of these contributions. These results are quite similar to ones obtained by Claveria et al. (2007) (Table 9). 5. Conclusion Surveys improve the forecasting performance by adding explanatory power of a model based on only past values of manufacturing growth. The question related to production expectations (Q5) improves the forecasting performance more than the headline index. It may be better for researchers to use not just the headline index, but also production-expectations data or the Economic Sentiment Indicator if their goal is to more-accurately predict manufacturing growth.

Similar results, which are not given here, are obtained for mean absolute errors. Surveys were discontinued in Ireland in April 2008. This fact may contribute to the result pertaining to that country. Bulgaria joined the EU in January 2007.
15

14

L.R. Klein, S. zmucur / Economic Modelling 27 (2010) 14531462

1457

Fig. 2. Manufacturing growth (y-o-y), Industrial Condence Indicator, and production expectations.

Table 3 Rank according to adjusted determination coefcient (R_bar 2), standard error of the regression (SEE), Akaike information criterion (AIC) and Schwarz criterion. Rank 0. AR (12) 1. Production 2. Orders 3. Export orders 4. Stocks 5. Production expectations 6. Price expectations 7. Employment expectations 8. Business Conditions Index 9. Industrial Condence Indicator 0 2 12 10. Economic Sentiment Indicator 6 15 19

R_bar 2 1 1 or 2 1, 2, or 3

1 1 1

3 6 10

0 0 0

0 1 1

2 2 2

11 15 17

0 2 3

2 5 6

3 7 13

Standard error of the regression 1 0 3 1 or 2 0 5 1, 2, or 3 0 9 Akaike information criterion 1 0 3 1 or 2 1 5 1, 2, or 3 1 9 Schwartz criterion 1 1 1 or 2 2 1, 2, or 3 3

0 0 1

0 1 2

1 1 2

10 15 17

1 4 5

1 4 5

4 8 14

0 3 12

8 15 17

0 0 1

0 1 2

1 1 2

10 15 18

1 3 3

2 3 5

3 8 14

0 4 12

8 15 17

3 4 8

0 0 0

0 1 2

1 1 2

10 15 18

1 3 3

2 3 5

3 8 14

0 4 12

7 15 17

1458

L.R. Klein, S. zmucur / Economic Modelling 27 (2010) 14531462

Table 4 Root mean square error for 1-period ahead forecasts (2007M012009M06). 0. AR (12) 1. Production 2. Orders 3. Export orders 3.06 2.62 6.51 3.70 3.58 3.46 5.24 5.19 2.58 3.30 3.59 4.84 9.30 3.05 4.00 8.39 6.76 1.83 4.21 3.58 3.45 6.45 4.53 3.46 1.99 1.66 2.20 1.96 4. Stocks 5. Production expectations 2.36 2.34 6.64 3.49 2.97 3.13 5.21 4.15 2.34 2.59 2.67 4.12 9.24 2.84 3.44 7.24 6.61 1.52 3.83 3.01 3.41 5.98 4.42 3.04 1.93 1.36 1.88 1.55 6. Price expectations 3.10 2.56 6.67 3.54 3.50 4.02 5.66 5.17 2.63 3.16 3.76 4.43 9.18 3.09 4.45 8.58 6.89 2.21 4.06 3.85 3.49 6.78 4.98 3.55 1.99 1.65 2.08 1.85 7. Employment expectations 3.00 2.49 6.43 3.94 3.94 2.96 4.96 4.68 2.58 3.44 3.10 5.00 9.56 3.07 3.90 8.32 6.89 2.30 4.15 3.46 3.40 6.32 4.20 3.41 2.12 1.58 2.17 1.92 8. Business Conditions Index 2.96 2.60 6.55 3.14 3.25 3.15 4.62 4.75 2.54 3.05 2.94 4.57 9.24 2.92 3.75 7.86 6.50 1.81 4.10 3.35 3.42 6.04 4.36 3.47 2.44 1.72 2.21 1.99 9. Industrial Condence Indicator 2.92 2.64 6.55 3.01 3.29 3.36 4.78 4.62 2.53 3.00 2.80 4.53 9.29 2.95 3.81 7.70 6.47 1.77 4.07 3.26 3.37 6.19 4.43 3.43 2.40 1.68 2.20 2.00 10. Economic Sentiment Indicator 2.61 2.44 6.50 3.17 2.99 3.58 4.80 4.42 2.54 2.97 2.80 4.59 8.97 2.98 4.13 7.47 6.29 1.42 4.24 3.16 3.41 6.33 4.05 2.99 2.10 1.35 2.09 1.75

Austria Belgium Bulgaria Cyprus Czech Republic Denmark Estonia Finland France Germany Greece Hungary Ireland Italy Latvia Lithuania Luxembourg Netherlands Poland Portugal Romania Slovakia Slovenia Spain Sweden UK Euro Area (EA16) European Union (EU27)

3.01 2.53 6.42 3.76 3.73 4.34 5.61 4.87 2.63 3.25 3.87 4.83 7.87 3.09 4.23 8.62 6.72 1.24 4.11 3.76 3.39 6.87 5.22 3.43 1.98 1.56 2.07 1.85

2.87 2.37 6.52 3.47 3.02 3.50 4.93 4.47 2.58 2.81 3.02 4.75 9.06 2.83 3.64 8.04 6.55 1.47 3.79 3.59 3.46 6.77 4.59 3.41 2.11 1.61 2.05 1.82

3.27 2.65 6.52 3.82 3.61 3.43 4.97 5.13 2.61 3.30 3.07 4.83 9.58 2.94 4.10 8.36 6.72 1.77 4.30 3.49 3.39 6.56 4.59 3.51 2.10 1.75 2.24 2.10

3.01 2.74 6.78 3.65 3.77 4.42 5.44 4.85 2.62 3.13 3.48 4.90 9.14 3.27 4.29 8.22 6.49 1.81 4.12 3.76 3.53 7.31 5.27 3.44 2.42 1.69 2.17 1.92

Table 5 Rank of models based on RMSE. Rank 0. AR (12) 1. Production 2. Orders 3. Export orders 4. Stocks 5. Production expectations 6. Price expectations 7. Employment expectations 8. Business Conditions Index 9. Industrial Condence Indicator 2 6 10 10. Economic Sentiment Indicator 4 13 18

1-period ahead forecast 1 3 2 1 or 2 5 6 1, 2, or 3 7 12 2-period ahead forecast 1 3 3 1 or 2 5 7 1, 2, or 3 5 13 3-period ahead forecast 1 3 4 1 or 2 4 7 1, 2, or 3 5 12

0 0 1

0 0 0

0 0 1

15 20 20

0 1 3

1 3 4

1 3 8

0 0 1

0 0 1

0 0 0

15 20 20

0 1 3

1 3 5

1 2 10

2 5 9

3 14 17

0 0 1

0 0 0

0 0 0

13 20 22

0 1 4

1 3 4

2 5 11

2 5 9

3 12 16

L.R. Klein, S. zmucur / Economic Modelling 27 (2010) 14531462

1459

Fig. 3. Manufacturing growth (y-o-y), actual and one-period ahead ex ante forecasts based on the benchmark model (AR (12)), and production expectations.

Fig. 4. Manufacturing growth (y-o-y), actual and two-period ahead ex ante forecasts based on the benchmark model (AR (12)), and production expectations.

1460

L.R. Klein, S. zmucur / Economic Modelling 27 (2010) 14531462

Table 6 Correlation between actual and 1-period ahead forecasts. Correlation 0. AR (12) 1. Production 2. Orders 3. Export orders 0.961 0.976 0.912 0.859 0.967 0.937 0.972 0.951 0.946 0.966 0.856 0.952 0.258 0.964 0.943 0.763 0.865 0.704 0.894 0.882 0.943 0.938 0.950 0.952 0.988 0.964 0.980 0.982 4. Stocks 5. Production expectations 0.974 0.978 0.891 0.881 0.975 0.943 0.970 0.955 0.959 0.981 0.905 0.957 0.224 0.967 0.958 0.859 0.873 0.700 0.906 0.895 0.944 0.950 0.945 0.964 0.991 0.977 0.992 0.992 6. Price expectations 0.958 0.974 0.886 0.870 0.965 0.899 0.965 0.946 0.938 0.963 0.856 0.949 0.221 0.959 0.930 0.735 0.855 0.720 0.900 0.850 0.941 0.933 0.933 0.952 0.988 0.963 0.977 0.979 7. Employment expectations 0.962 0.978 0.907 0.843 0.971 0.947 0.973 0.953 0.945 0.970 0.870 0.933 0.239 0.959 0.951 0.792 0.849 0.724 0.899 0.867 0.945 0.942 0.956 0.952 0.989 0.966 0.979 0.982 8. Business Conditions Index 0.966 0.978 0.904 0.893 0.972 0.948 0.977 0.957 0.949 0.978 0.885 0.951 0.284 0.967 0.950 0.802 0.875 0.698 0.908 0.884 0.943 0.949 0.956 0.955 0.978 0.969 0.985 0.986 9. Industrial Condence Indicator 0.966 0.977 0.903 0.904 0.971 0.948 0.975 0.956 0.949 0.977 0.895 0.949 0.266 0.967 0.949 0.830 0.875 0.687 0.907 0.885 0.944 0.944 0.956 0.955 0.979 0.971 0.985 0.987 10. Economic Sentiment Indicator 0.974 0.978 0.907 0.889 0.974 0.940 0.973 0.957 0.946 0.969 0.895 0.948 0.308 0.965 0.948 0.813 0.880 0.715 0.902 0.881 0.944 0.943 0.954 0.964 0.990 0.977 0.986 0.988

Austria Belgium Bulgaria Cyprus Czech Republic Denmark Estonia Finland France Germany Greece Hungary Ireland Italy Latvia Lithuania Luxembourg Netherlands Poland Portugal Romania Slovakia Slovenia Spain Sweden UK EA16 EU27

0.952 0.970 0.914 0.861 0.966 0.890 0.963 0.939 0.941 0.960 0.865 0.941 0.363 0.957 0.935 0.800 0.858 0.706 0.896 0.841 0.944 0.933 0.922 0.952 0.988 0.965 0.977 0.980

0.966 0.977 0.903 0.864 0.974 0.922 0.970 0.952 0.948 0.974 0.877 0.944 0.286 0.966 0.950 0.818 0.867 0.704 0.922 0.860 0.941 0.932 0.944 0.958 0.989 0.967 0.987 0.987

0.963 0.976 0.905 0.850 0.967 0.944 0.971 0.952 0.945 0.971 0.880 0.942 0.296 0.966 0.942 0.797 0.863 0.677 0.896 0.871 0.944 0.936 0.948 0.953 0.988 0.967 0.981 0.983

0.952 0.969 0.896 0.900 0.966 0.894 0.967 0.943 0.942 0.967 0.859 0.941 0.247 0.955 0.933 0.812 0.868 0.678 0.896 0.840 0.945 0.924 0.923 0.952 0.981 0.964 0.979 0.982

Table 7 Rank based on correlation between actual and ex ante forecasts. Rank 0. AR (12) 1. Production 2. Orders 3. Export orders 4. Stocks 5. Production expectations 6. Price expectations 7. Employment expectations 8. Business Conditions Index 9. Industrial Condence Indicator 2 7 15 10. Economic Sentiment Indicator 3 11 14

1-period ahead forecasts 1 2 1 or 2 2 1, 2, or 3 2

1 2 7

0 0 1

0 2 2

0 2 2

15 17 17

0 1 1

2 3 8

3 10 15

2-period ahead forecasts 1 2 1 1 or 2 2 4 1, 2, or 3 2 11 3-period ahead forecasts 1 2 2 1 or 2 2 5 1, 2, or 3 2 12

0 0 2

0 2 2

1 1 1

13 17 17

1 1 1

0 1 6

6 12 15

1 7 12

3 10 15

0 0 0

0 3 4

1 1 1

12 15 17

0 0 1

0 2 5

7 10 16

1 8 11

3 11 15

L.R. Klein, S. zmucur / Economic Modelling 27 (2010) 14531462 Table 8 DieboldMariano statistics. 1. Production 2. Orders 3. Export orders 0.38 1.04 0.20 0.93 1.04 1.47 3.23 0.83 1.05 0.06 0.55 0.86 0.88 0.69 1.40 0.68 0.30 0.71 1.04 1.53 0.55 0.94 3.34 1.14 1.20 4. Stocks 5. Production expectations 1.11 1.26 0.74 1.00 1.42 1.21 1.65 1.91 1.20 1.86 1.32 0.65 1.32 1.65 1.17 0.30 1.07 0.17 1.37 1.37 1.53 0.61 0.90 0.76 1.11 6. Price expectations 1.06 1.17 0.57 2.25 1.26 0.84 0.73 0.01 1.31 0.45 1.29 0.74 0.12 1.01 0.34 1.31 0.44 0.16 1.42 1.12 1.61 2.17 0.29 2.36 0.91 0.39 7. Employment expectations 0.07 0.53 1.25 0.93 1.06 1.56 1.05 1.06 1.33 1.51 1.20 1.10 0.93 1.09 1.35 0.30 1.00 0.14 2.37 1.56 0.38 9.65 0.82 0.82 0.70 8. Business Conditions Index 0.17 1.16 0.88 1.70 1.21 1.17 2.02 1.70 0.88 1.42 0.93 0.41 1.87 1.11 1.02 0.92 0.03 0.70 0.72 1.04 1.68 1.18 2.49 2.53 9. Industrial Condence Indicator 0.31 1.38 0.84 2.28 1.26 1.14 1.88 1.79 1.09 1.71 1.08 0.72 2.01 1.12 1.07 0.83 0.16 0.94 0.36 1.27 2.12 0.05 1.22 1.22 2.19

1461

10. Economic Sentiment Indicator 0.55 0.83 1.48 1.65 1.21 1.88 1.44 1.09 1.71 1.12 0.64 2.38 0.57 1.01 0.98 0.50 1.33 0.22 0.96 1.33 1.51 7.02 0.86 0.09 0.72

Austria Belgium Bulgaria Cyprus Czech Republic Denmark Estonia France Germany Greece Hungary Ireland Italy Latvia Lithuania Luxembourg Poland Portugal Romania Slovakia Slovenia Spain Sweden UK EA16 EU27

0.50 1.14 0.73 1.17 1.44 1.22 2.69 0.72 1.21 1.37 0.85 0.18 1.46 1.44 1.03 1.05 2.69 0.42 1.41 0.22 1.33 2.24 2.66 1.34 0.10 0.27

1.20 5.75 0.79 0.28 0.71 0.99 2.01 1.01 1.41 0.17 1.27 2.14 0.65 0.86 0.01 0.69 0.52 0.43 0.84 2.61 2.26 4.04 1.48 2.06

0.72 1.45 0.95 0.35 1.94 0.91 1.22 1.14 1.49 1.12 0.39 0.04 0.94 0.73 0.94 1.24 0.83 0.84 0.82 1.64 0.31 1.27 7.44 1.46 1.31

Table 9 Rank according to DieboldMariano statistics. Rank 1. Production 2. Orders 3. Export orders 4. Stocks 5. Production expectations 6. Price expectations 7. Employment expectations 8. Business Conditions Index 0 1 3 9. Industrial Condence Indicator 2 4 10 10. Economic Sentiment Indicator 4 8 14

1-period ahead forecast, square errors 1 4 1 1 or 2 11 2 1, 2, or 3 14 3 1-period ahead forecast, absolute errors 1 2 0 1 or 2 7 1 1, 2, or 3 11 3 2-period ahead forecast, square errors 1 2 1 1 or 2 9 2 1, 2, or 3 13 4 2-period ahead forecast, absolute errors 1 4 2 1 or 2 7 3 1, 2, or 3 13 3 3-period ahead forecast, square errors 1 2 2 1 or 2 6 4 1, 2, or 3 11 4 3-period ahead forecast, absolute errors 1 4 0 1 or 2 7 2 1, 2, or 3 12 3

1 1 2

2 2 3

10 15 18

1 6 6

1 3 5

1 4 4

1 2 2

10 13 15

5 6 8

1 2 5

0 1 6

2 6 8

4 11 16

1 1 2

2 2 3

12 17 18

1 4 7

1 2 4

0 0 4

2 8 11

4 8 12

0 2 3

1 2 3

7 16 19

2 3 5

1 3 5

2 2 3

3 5 8

5 9 16

1 1 2

2 2 4

12 16 18

1 3 6

1 1 3

0 2 5

3 9 13

2 9 12

2 2 3

1 2 3

9 16 20

1 3 6

1 3 5

0 1 2

5 7 12

4 9 12

1462

L.R. Klein, S. zmucur / Economic Modelling 27 (2010) 14531462 Eurostat, 2009. Industry production Index-Monthly Data (2005 = 100) (NACE Rev 2), October 2009. http://epp.eurostat.ec.europa.eu/portal/page/portal/statistics/ search_database. Hansson, J., Jansson, P., Lf, M., 2005. Business survey data: do they help in forecasting GDP growth? International Journal of Forecasting 21 (2), 377389 AprilJune 2005. Kauppi, E., Lassila, J., Tersvirta, T., 1996. Short-term forecasting of industrial production with business survey data: experience from Finland's great depression 19901993. International Journal of Forecasting 12 (3), 373381 September 1996. Klein, L.R., zmucur, S., 2002a. Consumer behavior under the inuence of terrorism within the United States. The Journal of Entrepreneurial Finance & Business Ventures 7 (3), 115 Fall 2002. Klein, L.R., zmucur, S., 2002b. The predictive power of survey results in macroeconomic analysis. In: Welfe, Wladyslaw (Ed.), Macromodels 2001. University of Lodz, Poland, pp. 181197. 2002. Klein, L.R., zmucur, S., 2004. Some possibilities for indicator analysis in economic forecasting. In: Dua, Pami (Ed.), Business Cycles and Economic Growth: An Analysis Using Leading Indicators. Oxford University Press, pp. 243257. 2004. Lemmens, A., Croux, C., Dekimpe, M.G., 2005. On the predictive content of production surveys: a pan-European study. International Journal of Forecasting 21 (2), 363375 AprilJune 2005.

References
Claveria, O., Pons, E., Ramos, R., 2007. Business and consumer expectations and macroeconomic forecasts. International Journal of Forecasting 23 (1), 4769 JanuaryMarch 2007. Deitz, R., Steindel, C., 2005. The predictive abilities of the New York Fed's Empire State Manufacturing Survey. Current Issues in Economics and Finance Second District Highlights, January 2005, Volume 11, Number 1. http://www.ny.frb.org/research/ current_issues/ci11-1.pdf. European Commission Directorate-General for Economic Financial Affairs, 2006. European Economy, Special Report No 5 The Joint Harmonised EU Programme of Business and Consumer Surveys, December 2006. http://ec.europa.eu/economy_nance/publications/publication7568_en.pdf. European Commission Directorate-General for Economic Financial Affairs, 2007. The Joint Harmonised EU Programme of Business and Consumer Surveys, User Guide, Updated 4 July 2007. http://ec.europa.eu/economy_nance/indicators/business_consumer_surveys/user guide_en.pdf. European Commission Directorate-General for Economic Financial Affairs, 2009a. Business and Consumer Surveys, 29 October 2009. http://ec.europa.eu/economy_nance/db_indicators/surveys9185_en.htm. European Commission Directorate-General for Economic Financial Affairs, 2009b. Business and Consumer Surveys, Workshop Documents, 14 October 2009. http:// ec.europa.eu/economy_nance/db_indicators/surveys9217_en.htm.