European Commission DG ECFIN Directorate General Economic and Financial Affairs A new survey-based indicator to track Industrial Production Olivier Biau – Angela D’Elia Fourth joint EU-OECD.
Download ReportTranscript European Commission DG ECFIN Directorate General Economic and Financial Affairs A new survey-based indicator to track Industrial Production Olivier Biau – Angela D’Elia Fourth joint EU-OECD.
European Commission DG ECFIN Directorate General Economic and Financial Affairs A new survey-based indicator to track Industrial Production Olivier Biau – Angela D’Elia Fourth joint EU-OECD workshop on business and consumer opinion surveys Brussels, 13 October 2009 ECFIN– 13/10/2009 Slide 1 Outline DG ECFIN … during the crisis several indicator-based models performed in a less satisfactory way … • Are there alternatives to balance statistics? • Where and how much information in detailed answers? • A new synthetic indicator in manufacturing to track and forecast IP growth • In-sample and out-of-sample performance • Further developments Olivier Biau – Angela D’Elia European Commission, Economic and Financial Affairs ECFIN – 13/10/2009 Slide 2 Introduction DG ECFIN • Survey data – Barometer of the economic situation – Gauge the economic stance – … but they are qualitative • Balance statistics (difference between % of respondents giving positive and negative replies) – Easy to implement, read and follow (no revisions) – Highly correlated with hard aggregates – … but based on restrictive assumptions (Pesaran&Weale, 2006) Olivier Biau – Angela D’Elia European Commission, Economic and Financial Affairs ECFIN – 13/10/2009 Slide 3 Introduction & Rationale DG ECFIN INDUSTRY INDUSTRY euro area INDUSTRY---euro euroarea area 25 65 8080 15 45 5 5 7070 6060 25 -5 5050 -15 -15 5 4040 -25 -25 -15 -35 -35 3030 -35 -45 -45 1010 -55 -55 -55 00 Jan-09 Jan-09 Jan-09 Jan-07 Jan-07 Jan-07 Jan-05 Jan-05 Jan-05 Jan-03 Jan-03 Jan-03 Jan-01 Jan-01 Jan-01 Jan-99 Jan-99 Jan-99 Jan-97 Jan-97 Jan-97 Jan-95 Jan-95 Jan-95 Jan-93 Jan-93 Jan-93 Jan-91 Jan-91 Jan-91 Jan-89 Jan-89 Jan-89 Jan-87 Jan-87 Jan-87 Jan-85 Jan-85 Jan-85 2020 Balance (nsa) Q1 Balance (nsa) Q1 Stable (nsa) Q1 -rhs Balance (nsa) Q1 Negative (nsa) Q1 -rhs Stable (nsa) Q1 -rhs Evidence of moving from « do not change » to « negative » answers Olivier Biau – Angela D’Elia European Commission, Economic and Financial Affairs ECFIN – 13/10/2009 Slide 4 DG ECFIN Where and how much information in detailed answers? • Principal components analysis (Hild, 2003; Etter et al, 2004) – to find the « best » linear combination of 3 answer modalities • Regression analysis – to identify the combination with highest explanatory power wrt hard reference series (y-o-y Industrial Production growth in EA) Dataset: Industry Business&Consumer Survey data (Jan91 –Jun09) Industrial Production Index (Jan91 – Apr09) Olivier Biau – Angela D’Elia European Commission, Economic and Financial Affairs ECFIN – 13/10/2009 Slide 5 Industry survey –monthly questions DG ECFIN • Q1 How has your production developed over the past 3 months? It has... + increased = remained unchanged − decreased • Q2 Do you consider your current overall order books to be...? + more than sufficient (above normal) = sufficient (normal for the season) − not sufficient (below normal) • Q3 Do you consider your current export order books to be...? + more than sufficient (above normal) = sufficient (normal for the season) − not sufficient (below normal) • Q4 Do you consider your current stock of finished products to be...? + too large (above normal) = adequate (normal for the season) − too small (below normal) • Q5 How do you expect your production to develop over the next 3 months? It will... + increase = remain unchanged − decrease Olivier Biau – Angela D’Elia European Commission, Economic and Financial Affairs ECFIN – 13/10/2009 Slide 6 Principal Components analysis DG ECFIN Explained PC1 Loadings variance Q1 %+ (P) 0.56 %= (S) 0.55 %– (M) -0.62 87% Q2 0.56 0.57 -0.60 92% Q3 0.54 0.57 -0.62 87% Q4 -0.70 0.64 0.30 68% Q5 0.57 0.54 -0.62 85% PC1 = 0.55*P + 0.55*S – 0.6*M = 0.55*(P + S) – 0.62*M = 0.55* (100 – M) – 0.62*M PC1 = a + b*M All the relevant information is contained in the % of respondents who give a negative answer Olivier Biau – Angela D’Elia European Commission, Economic and Financial Affairs ECFIN – 13/10/2009 Slide 7 Regression analysis DG ECFIN IPt = c1 *Pt + c2 *St + c3 *Mt + ut • M modality (negative answers) is always the most significant for explaining IP, across all questions • Model built on Q1 (past production) and Q5 (expected production) has the best fit (highest R2) • For Q1, Q2 and Q5, the null hypothesis c1=c2=c cannot be rejected IPt = c* (Pt +St)+ c3 *Mt + ut = c* (100 - Mt)+ c3 *Mt + ut = α + β* Mt + ut Focus on % of negative answers to Q1, Q2 and Q5 Olivier Biau – Angela D’Elia European Commission, Economic and Financial Affairs ECFIN – 13/10/2009 Slide 8 A new indicator to track Industrial Production DG ECFIN Out of the negative modality of answer to Q1, Q2 and Q5, it is possible to build 7 different indicators Olivier Biau – Angela D’Elia European Commission, Economic and Financial Affairs 0 -1.0 -5 -2.0 -10 -3.0 y-o-y growth 0.0 -15 M15 IP (RHS) ECFIN – 13/10/2009 Slide 9 2009M01 2008M01 2007M01 2006M01 2005M01 2004M01 2003M01 2002M01 2001M01 2000M01 1999M01 1998M01 -25 1997M01 -5.0 1996M01 -20 1995M01 -4.0 1994M01 2 7 3 6 1 5 4 5 1993M01 -0.946 -0.817 -0.937 -0.900 -0.954 -0.923 -0.935 1.0 1992M01 M1 M2 M5 M12 M15 M25 M125 10 1991M01 Indicator Correlation Ranking with IP 2.0 DG ECFIN A comparison with Industrial Confidence Indicator and Business Climate Indicator • Correlation with Industrial Production Index Indicator Tn = Correlation with IP Tn M15 -0.954 ICI 0.878 -5.33 BCI 0.871 -5.66 arctanh(corr ) arctanh(corr ) a 1 2 ~ N(0,1) 2 n 3 Test of equality of correlation coefficients M15 correlation with IP is significantly higher than those of ICI and BCI Olivier Biau – Angela D’Elia European Commission, Economic and Financial Affairs ECFIN – 13/10/2009 Slide 10 In-sample analysis M15 or ICI or BCI DG ECFIN The model: Indicator R² IPt = a0+ a1 IPt-1 + a2 Xt + a3 Xt-1 + ut S.E. of regression Durbin Watson h modified AIC DW M15 0.93 1.16 2.06 -1.39 3.15 ICI 0.89 1.43 2.54 -6.82 3.23 BCI 0.89 1.42 2.62 -7.67 3.21 10 5 0 -5 -10 -15 -20 -25 92 94 96 98 00 Obs erved IP Olivier Biau – Angela D’Elia European Commission, Economic and Financial Affairs 02 04 06 08 Fitted IP ECFIN – 13/10/2009 Slide 11 Out-of-sample analysis (M15 vs ICI and BCI) DG ECFIN • One month horizon forecast • Sample periods: Jan05 – Apr09 and Jan07 – Apr09 Jan07 – Apr09 Jan05 – Apr09 Indicator MAE MSE M15 1.19 2.17 ICI 1.59 4.49 BCI 1.66 4.62 Indicator MAE MSE M15 1.39 2.90 -2.57 ICI 2.04 6.90 -2.50 -2.97 BCI 2.07 6.77 -2.66 Harvey et al test Harvey et al test Harvey et al test – in small sample- the null hypothesis of no difference between the MSEs of 2 models M15 is significantly better in forecasting IP Olivier Biau – Angela D’Elia European Commission, Economic and Financial Affairs ECFIN – 13/10/2009 Slide 12 Out-of-sample analysis (M15 vs ICI and BCI) DG ECFIN 10 5 0 -5 -10 -15 -20 IP M15_FORECAST ICI_FORECAST BCI_FORECAST -25 2007 2008 • M15 was the first to signal a negative yoy growth for IP (summer 2008) • M15 correctly tracked the IP during the crisis, and predicted the improvement in May09 Olivier Biau – Angela D’Elia European Commission, Economic and Financial Affairs ECFIN – 13/10/2009 Slide 13 Conclusions and further developments DG ECFIN • Better performance of M15 in tracking and forecasting Industrial Production growth, especially during last crisis • … but M15 is not necessarily a confidence indicator To build a complementary confidence indicator as a common factor over the negative answers to all the 5 questions of industry survey To apply same approach to other surveys of BCS Olivier Biau – Angela D’Elia European Commission, Economic and Financial Affairs ECFIN – 13/10/2009 Slide 14