Executive+Popularity+in+France+Version

Executive Popularity in France: The Promise and Pitfalls of Time Series Data Research and Methods Symposium, October 1st 2004 The problems  The substantive problem: how do macroeconomic conditions affect support for the dual executive (president and prime minister) in France?  The methodological problem:  What techniques are best suited to modeling timeseries data?  Do any of these models have a reliable predictive component (forecasting) Some hypothetical data series Stationary series Y(t) = α, where the estimated constant α is the sample mean Linear trend Y(t) = α + β(t), where α is the intercept and β the slope of the trend line “Random walk” Y(t) = Y(t-1) + α, where α is the mean of the first difference (i.e. average change from one period to the next) France: Executive Popularity France: the dual executive  In de Gaulle‟s formulation, the president has responsibility for „high politics‟, while the prime minister is responsible for „day to day affairs‟  The president is directly elected (since 1965), for a five year term (since 2002, seven years previously)  The president appoints the prime minister, who is „responsible‟ to parliament  The president has no constitutional authority to fire the prime minister, but has acquired the de facto capacity to do this  The president may dismiss the national assembly and call for new elections (no more than once a year)  The possibility exists that the president and prime minister may be drawn from different ideological camps (cohabitation)  Cohabitation has occurred three times: 1986-88 (Mitterrand/Chirac), 1993-5 (Mitterrand/Balladur), and 1997-2002 (Chirac/Jospin). Executive popularity in France • Lewis-Beck (1980) finds that Prime Ministers suffer a greater decline in popularity than Presidents due to negative effects of inflation and unemployment • Hibbs (1981) finds negative effects of unemployment on presidential approval (but positive effect of inflation!) • Appleton (1986) finds negative relationship between unemployment and both presidential and prime ministerial approval, but no inflation effect • Anderson (1995) suggests that the relationship is more complex, and depends upon the ideological placement of the prime minister vis a vis the president • Lecaillon (1980) finds no impact of macroeconomic variables on executive popularity • Anonymous (2004) finds that: presidential approval linked to unemployment, prime ministerial approval suffers from higher inflation, and that presidential popularity rises during cohabitation Executive Popularity Indicators in France IFOP (Since 1958, aperiodic until 1968, then monthly): “Are you satisfied with [name] as [President, Prime Minister] of France?” June 2004: Chirac 45% Raffarin 32% SOFRES (Since 1978): “How reliable do you think [name] is in dealing with France‟s current problems?” (1974-78): “How effective do you think [name] is in dealing with France‟s current problems?” June 2004 Chirac 35% Raffarin 28% Also BULL-BVA series (1982-1990‟s) Presidential Popularity in France, 1978-2004 80 70 60 50 40 30 20 Date Giscard d‟Estaing 04 20 1/ 2 /0 00 02 1 /2 01 /0 0 10 1 /2 00 /0 0 06 1 /2 98 /0 9 02 1 /1 97 /0 9 10 1 /1 96 /0 9 06 1 /1 94 /0 9 02 1 /1 93 /0 9 10 1 /1 92 /0 9 06 1 /1 90 /0 9 02 1 /1 89 /0 9 10 1 /1 88 /0 9 06 1 /1 86 /0 9 02 1 /1 85 /0 9 10 1 /1 84 /0 9 06 1 /1 82 /0 9 02 1 /1 81 /0 9 10 1 /1 80 /0 9 06 1 /1 78 /0 9 02 1 /1 /0 10 Source: SOFRES President Yes Mitterrand Chirac Prime Ministerial Popularity in France, 1978-2004 7. Bérégovoy 8. Balladur 6. Cresson 80 70 60 50 40 30 20 10 Date PM Yes 1 04 20 1/ 2 /0 00 02 1 /2 01 /0 0 10 1 /2 00 /0 0 06 1 /2 98 /0 9 02 1 /1 97 /0 9 10 1 /1 96 /0 9 06 1 /1 94 /0 9 02 1 /1 93 /0 9 10 1 /1 92 /0 9 06 1 /1 90 /0 9 02 1 /1 89 /0 9 10 1 /1 88 /0 9 06 1 /1 86 /0 9 02 1 /1 85 /0 9 10 1 /1 84 /0 9 06 1 /1 82 /0 9 02 1 /1 81 /0 9 10 1 /1 80 /0 9 06 1 /1 78 /0 9 02 1 /1 /0 10 Source: SOFRES 10 11 8 4 5 2 3 6 7 9 11. Raffarin 2. Mauroy 5. Rocard 3. Fabius 10. Jospin 4. Chirac 9. Juppé 1. Barre Executive Popularity in France, 1978-2004 President Yes 80 70 60 50 40 30 20 10 Date 04 20 1 / 02 /0 20 02 1 / 001 /0 2 10 01 / 000 / 2 06 1 / 98 /0 19 02 1 / 997 /0 1 10 01 / 996 / 1 06 1 / 994 /0 1 02 01 / 993 / 1 10 1 / 992 /0 1 06 01 / 990 / 1 02 1 / 989 /0 1 10 1 / 88 /0 19 06 1 / 86 /0 19 02 1 / 985 /0 1 10 01 / 984 / 1 06 1 / 982 /0 1 02 01 / 981 / 1 10 1 / 980 /0 1 06 01 / 978 / 1 02 1 / /0 10 Source: SOFRES PM Yes Value Executive Popularity in France – Giscard d‟Estaing President Yes 70 60 50 40 30 20 Date 81 19 1/ /0 81 04 /19 1 /0 80 02 /19 1 /0 80 12 /19 1 /0 80 10 /19 1 /0 80 08 /19 1 /0 80 06 /19 1 /0 80 04 /19 1 /0 79 02 /19 1 /0 79 12 /19 1 /0 79 10 /19 1 /0 79 08 /19 1 /0 79 06 /19 1 /0 79 04 /19 1 /0 78 02 /19 1 /0 78 12 /19 1 /0 10 Source: SOFRES PM Yes Executive Popularity in France – Mitterrand President Yes 80 70 60 50 40 30 20 10 Date 94 19 1/ 4 /0 99 12 1 /1 3 /0 99 03 1 /1 2 /0 99 06 1 /1 1 /0 99 09 1 /1 1 /0 99 12 1 /1 0 /0 99 03 1 /1 9 /0 98 06 1 /1 8 /0 98 09 1 /1 8 /0 98 12 1 /1 7 /0 98 03 1 /1 6 /0 98 06 1 /1 5 /0 98 09 1 /1 5 /0 98 12 1 /1 4 /0 98 03 1 /1 3 /0 98 06 1 /1 2 /0 98 09 1 /1 2 /0 98 12 1 /1 1 /0 98 03 1 /1 /0 06 Source: SOFRES PM Yes Balladur Chirac Executive Popularity in France – Chirac President Yes 80 70 60 50 40 30 20 Date 04 20 1/ 3 /0 00 06 1 /2 3 /0 00 12 1 /2 2 /0 00 06 1 /2 2 /0 00 12 1 /2 1 /0 00 06 1 /2 1 /0 00 12 1 /2 0 /0 00 06 1 /2 0 /0 00 12 1 /2 9 /0 99 06 1 /1 9 /0 99 12 1 /1 8 /0 99 06 1 /1 8 /0 99 12 1 /1 7 /0 99 06 1 /1 7 /0 99 12 1 /1 6 /0 99 06 1 /1 6 /0 99 12 1 /1 5 /0 99 06 1 /1 5 /0 99 12 1 /1 /0 06 Source: SOFRES Jospin PM Yes Relating executive popularity to macroeconomic conditions Inflation and Unemployment in France, 1978-2004 30 20 10 0 -10 Date 04 20 1/ 2 /0 00 02 1 /2 01 /0 0 10 1 /2 00 /0 0 06 1 /2 98 /0 9 02 1 /1 97 /0 9 10 1 /1 96 /0 9 06 1 /1 94 /0 9 02 1 /1 93 /0 9 10 1 /1 92 /0 19 06 01 / 90 / 9 02 1 /1 89 /0 9 10 1 /1 88 /0 9 06 1 /1 86 /0 9 02 1 /1 85 /0 9 10 1 /1 84 /0 9 06 1 /1 82 /0 9 02 1 /1 81 /0 9 10 1 /1 80 /0 9 06 1 /1 78 /0 9 02 1 /1 /0 10 Source: INSEE Giscard Mitterrand Chirac Unemployment Inflation Bivariate correlations of presidential and prime ministerial popularity, inflation, and unemployment Correlations Unemplo President Yes PM Yes Inflation yment 1 .404** .211** -.388** . .000 .000 .000 311 308 292 299 .404** 1 -.293** .201** .000 . .000 .001 308 308 289 296 .211** -.293** 1 -.652** .000 .000 . .000 292 289 292 282 -.388** .201** -.652** 1 .000 .001 .000 . 299 296 282 300 President Yes PM Yes Inflation Unemployment Pearson Correlation Sig. (2-tailed) N Pearson Correlation Sig. (2-tailed) N Pearson Correlation Sig. (2-tailed) N Pearson Correlation Sig. (2-tailed) N **. Correlation is significant at the 0.01 level (2-tailed). Simple OLS predicting presidential popularity Model Summary Model 1 R R Square .458 a .210 Adjusted R Square .199 Std. Error of the Estimate 7.979 a. Predictors: (Constant), Unemployment, Mitterrand, Inflation, Chirac Coefficientsa Unstandardized Coefficients B Std. Error 74.613 4.272 8.624 2.327 9.919 2.641 -.094 .140 -3.424 .438 Standardized Coefficients Beta .476 .523 -.050 -.617 Model 1 (Constant) Mitterrand Chirac Inflation Unemployment t 17.464 3.706 3.756 -.667 -7.825 Sig. .000 .000 .000 .505 .000 a. Dependent Variable: President Yes Extended OLS predicting presidential popularity Model Summary Model 1 R R Square a .697 .486 Adjusted R Square .473 Std. Error of the Estimate 6.475 a. Predictors: (Constant), Cohabitation, Time in Office (President), Unemployment, PM Yes, Mitterrand, Inflation, Chirac Coefficientsa Unstandardized Coefficients B Std. Error 68.351 4.070 -.696 2.106 -5.685 2.627 -.341 .127 -2.974 .372 .365 .038 -.079 .012 2.511 1.021 Standardized Coefficients Beta -.038 -.300 -.183 -.538 .508 -.394 .138 Model 1 (Constant) Mitterrand Chirac Inflation Unemployment PM Yes Time in Office (President) Cohabitation t 16.793 -.331 -2.164 -2.681 -7.988 9.663 -6.686 2.459 Sig. .000 .741 .031 .008 .000 .000 .000 .015 a. Dependent Variable: President Yes Extended OLS predicting presidential popularity with lagged dependent variable as predictor Model Summary Model 1 R R Square .919 a .844 Adjusted R Square .840 Std. Error of the Estimate 3.514 a. Predictors: (Constant), President Yes lagged, Cohabitation, Mitterrand, Inflation, Time in Office (President), PM Yes, Unemployment, Chirac Coefficientsa Unstandardized Coefficients B Std. Error 13.117 3.122 -1.308 1.149 -2.945 1.432 -.137 .070 -.525 .226 .108 .023 -.021 .007 .889 .563 .794 .031 Standardized Coefficients Beta -.073 -.158 -.074 -.096 .151 -.103 .050 .805 Model 1 (Constant) Mitterrand Chirac Inflation Unemployment PM Yes Time in Office (President) Cohabitation President Yes lagged t 4.202 -1.138 -2.056 -1.963 -2.326 4.686 -3.004 1.579 25.246 Sig. .000 .256 .041 .051 .021 .000 .003 .116 .000 a. Dependent Variable: President Yes Alternative extended OLS predicting presidential popularity with lagged dependent variable as predictor Model Summary Model 1 R .921 R Square .848 Adjusted R Square .843 Std. Error of the Estimate 3.480 Coefficientsa Unstandardized Coefficients B Std. Error 16.909 3.885 .782 .570 -.748 .729 .791 .032 -.030 .008 -2.261 1.327 -4.709 1.715 .119 .023 -.679 .245 -.323 .107 Standardized Coefficients Beta .044 -.027 .804 -.153 -.126 -.254 .166 -.119 -.152 Model 1 (Constant) Cohabitation Post-election President Yes lagged Time in Office (President) Mitterrand Chirac PM Yes PMA(UNEMPLOY,3) PMA(INFLAT,3) t 4.353 1.371 -1.026 24.774 -3.897 -1.704 -2.745 5.072 -2.773 -3.021 Sig. .000 .171 .306 .000 .000 .090 .006 .000 .006 .003 a. Dependent Variable: President Yes Extended OLS predicting prime ministerial popularity with lagged dependent variable as predictor Model Summary Model 1 R .961 R Square .923 Adjusted R Square .918 Std. Error of the Estimate 3.548 Coefficientsa Unstandardized Coefficients B Std. Error 29.199 5.282 .155 .074 -3.553 .460 .508 .048 .221 .035 -.413 .032 15.567 1.413 29.161 2.170 19.828 2.238 26.562 2.129 10.758 1.843 30.285 2.304 44.613 2.955 24.909 2.563 40.557 2.687 25.238 2.120 Standardized Coefficients Beta .060 -.464 .364 .219 -.447 .428 .580 .459 .704 .177 .478 1.050 .566 1.349 .341 Model 1 (Constant) Inflation Unemployment President Yes PM Yes lagged Time in office (PM) Mauroy Fabius Chirac (PM) Rocard Cresson Beregovoy Balladur Juppe Jospin Raffarin t 5.528 2.088 -7.722 10.674 6.282 -13.040 11.020 13.438 8.858 12.475 5.838 13.147 15.098 9.718 15.096 11.902 Sig. .000 .038 .000 .000 .000 .000 .000 .000 .000 .000 .000 .000 .000 .000 .000 .000 a. Dependent Variable: PM Yes Alternative extended OLS predicting prime ministerial popularity with lagged dependent variable as predictor Model Summary Model 1 R .964 R Square .929 Adjusted R Square .925 Coefficientsa Unstandardized Coefficients B Std. Error 40.486 5.889 -4.467 .482 -.010 .123 .513 .045 -.472 .033 .200 .034 14.785 1.369 29.287 2.116 20.088 2.204 26.626 2.120 8.985 1.913 29.333 2.290 46.180 2.854 25.764 2.472 42.138 2.649 22.987 2.184 Standardized Coefficients Beta -.558 -.004 .370 -.516 .199 .411 .591 .471 .714 .150 .469 1.102 .593 1.417 .334 Std. Error of the Estimate 3.388 Model 1 (Constant) PMA(UNEMPLOY,3) PMA(INFLAT,3) President Yes Time in office (PM) PM Yes lagged Mauroy Fabius Chirac (PM) Rocard Cresson Beregovoy Balladur Juppe Jospin Raffarin t 6.875 -9.262 -.085 11.346 -14.311 5.910 10.803 13.841 9.114 12.558 4.696 12.808 16.180 10.421 15.905 10.527 Sig. .000 .000 .932 .000 .000 .000 .000 .000 .000 .000 .000 .000 .000 .000 .000 .000 a. Dependent Variable: PM Yes Predicted versus actual values from extended OLS model of presidential popularity ARIMA predicted 80 70 60 50 40 30 20 Date 04 20 1 / 02 /0 20 02 01 / 01 / 0 10 1 /2 00 /0 0 06 1 /2 98 /0 19 02 01 / 97 / 9 10 1 /1 96 /0 19 06 01 / 94 / 9 02 1 /1 93 /0 9 10 1 /1 92 /0 19 06 01 / 90 / 9 02 1 /1 89 /0 19 10 01 / 88 / 9 06 1 /1 86 /0 19 02 01 / 85 / 9 10 1 /1 84 /0 9 06 1 /1 82 /0 19 02 01 / 81 / 9 10 1 /1 80 /0 19 06 01 / 78 / 9 02 1 /1 /0 10 OLS predicted Predicted versus actual values from extended OLS model of presidential popularity (Giscard) 70 60 50 40 30 Date 81 19 1/ /0 81 04 /19 1 /0 80 02 /19 1 /0 80 12 /19 1 /0 80 10 /19 1 /0 80 08 /19 1 /0 80 06 /19 1 /0 80 04 /19 1 /0 79 02 /19 1 /0 79 12 /19 1 /0 79 10 /19 1 /0 79 08 /19 1 /0 79 06 /19 1 /0 79 04 /19 1 /0 78 02 /19 1 /0 78 12 /19 1 /0 10 Predicted value President Yes Predicted versus actual values from extended OLS model of presidential popularity (Mitterrand) 80 70 60 50 40 30 20 Date 94 19 1/ 4 /0 99 12 1 /1 3 /0 99 03 1 /1 2 /0 99 06 1 /1 91 /0 9 09 1 /1 91 /0 9 12 1 /1 0 /0 99 03 1 /1 9 /0 98 06 1 /1 88 /0 9 09 1 /1 88 /0 9 12 1 /1 7 /0 98 03 1 /1 6 /0 98 06 1 /1 85 /0 9 09 1 /1 85 /0 9 12 1 /1 4 /0 98 03 1 /1 3 /0 98 06 1 /1 82 /0 9 09 1 /1 82 /0 9 12 1 /1 1 /0 98 03 1 /1 /0 06 Predicted value President Yes Predicted versus actual values from extended OLS model of presidential popularity (Chirac) 70 60 50 40 30 Date 04 20 1/ 3 /0 00 06 1 /2 03 /0 0 12 1 /2 2 /0 00 06 1 /2 2 /0 00 12 1 /2 1 /0 00 06 1 /2 01 /0 0 12 1 /2 00 /0 0 06 1 /2 0 /0 00 12 1 /2 9 /0 99 06 1 /1 9 /0 99 12 1 /1 98 /0 9 06 1 /1 98 /0 9 12 1 /1 7 /0 99 06 1 /1 7 /0 99 12 1 /1 6 /0 99 06 1 /1 96 /0 9 12 1 /1 95 /0 9 06 1 /1 5 /0 99 12 1 /1 /0 06 Predicted value President Yes Moving from OLS to ARIMA The autoregressive integrated moving average model (ARIMA) The model incorporates:  The autoregressive term p, which is the order of the autoregressive component The number of differences d, which is used to discount trends over time The moving average term q, which is the moving average of the prediction error   To fit the model, we need to examine the autocorrelation and the partial autocorrelation functions (ACF and PACF) Differenced series for presidential popularity, 1978-2004 40 30 20 10 0 -10 -20 Date Transforms: difference (1) 04 20 1/ 2 /0 00 02 1 /2 01 /0 0 10 1 /2 00 /0 0 06 1 /2 98 /0 9 02 1 /1 97 /0 9 10 1 /1 96 /0 9 06 1 /1 94 /0 9 02 1 /1 93 /0 9 10 1 /1 92 /0 9 06 1 /1 90 /0 9 02 1 /1 89 /0 9 10 1 /1 88 /0 9 06 1 /1 86 /0 9 02 1 /1 85 /0 9 10 1 /1 84 /0 9 06 1 /1 82 /0 9 02 1 /1 81 /0 9 10 1 /1 80 /0 9 06 1 /1 /0 02 Differenced series for prime ministerial popularity, 1978-2004 60 40 20 0 -20 -40 Date Transforms: difference (1) 04 20 1/ 2 /0 00 02 1 /2 01 /0 0 10 1 /2 00 /0 0 06 1 /2 98 /0 9 02 1 /1 97 /0 9 10 1 /1 96 /0 9 06 1 /1 94 /0 9 02 1 /1 93 /0 9 10 1 /1 92 /0 9 06 1 /1 90 /0 9 02 1 /1 89 /0 9 10 1 /1 88 /0 9 06 1 /1 86 /0 9 02 1 /1 85 /0 9 10 1 /1 84 /0 9 06 1 /1 82 /0 9 02 1 /1 81 /0 9 10 1 /1 80 /0 9 06 1 /1 /0 02 PM Yes ACF and PACF plots for presidential popularity President Yes 1.0 1.0 President Yes .5 .5 0.0 0.0 -.5 Partial ACF -.5 Confidence Limits Confidence Limits ACF -1.0 1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 Coefficient -1.0 1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 Coefficient Lag Number Lag Number ACF and PACF plots for presidential popularity (1 difference) President Yes 1.0 1.0 President Yes .5 .5 0.0 0.0 -.5 Partial ACF Confidence Limits -.5 Confidence Limits ACF -1.0 1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 Coefficient -1.0 1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 Coefficient Lag Number Transforms: difference (1) Lag Number Transforms: difference (1) ARIMA (1,0,1) model predicting presidential popularity Number of residuals 279 Standard error 3.2791659 Log likelihood -724.93328 AIC 1469.8666 SBC 1506.1787 Analysis of Variance: DF Adj. Sum of Squares 269 2950.8636 Residual Variance 10.752929 Residuals Variable AR1 MA1 MITTERRA CHIRAC INFLAT UNEMPLOY PM YES Time in Office Cohabitation CONSTANT B SEB T-RATIO APPROX. PROB. .939614 .021920 42.865332 .00000000 .225668 .065157 3.463472 .00062027 2.522298 4.175431 .604081 .54629903 .621881 7.942852 .078294 .93765206 -.059998 .059933 -1.001086 .31768488 -2.521530 1.267940 -1.988682 .04774939 .320467 . 037488 8.548599 .00000000 -.143606 .034891 -4.115908 .00005133 .486023 1.525982 .318498 .75035403 66.649067 10.639860 6.264092 .00000000 Alternative ARIMA (1,0,1) model predicting presidential popularity Number of residuals 273 Standard error 3.2141376 Log likelihood -701.86056 AIC 1425.7211 SBC 1465.4253 Analysis of Variance: DF Adj. Sum of Squares Residuals 262 2732.3972 Variables in the Model: B AR1 .937443 MA1 .283545 MITTERRA -2.919352 CHIRAC -5.153437 TIO -.080188 COHAB .957904 POSTELEC 3.371065 PMYES .290923 UNEMPL_2 -1.903296 INFLAT_2 -.076970 CONSTANT 62.296313 Residual Variance 10.330681 SEB T-RATIO APPROX. PROB. .024247 38.661481 .00000000 .066458 4.266519 .00002776 4.294587 -.679775 .49724719 7.609262 -.677258 .49883955 .036839 -2.176729 .03039386 1.469957 .651655 .51519511 .881233 3.825394 .00016328 .037170 7.826751 .00000000 1.265069 -1.504500 .13365712 .141287 -.544777 . 58637056 11.302350 5.511802 .00000008 Predicted versus actual values of presidential popularity from ARIMA model 80 70 60 50 40 30 20 Date 04 20 1/ 2 /0 00 02 1 /2 01 /0 20 10 01 / 00 / 0 06 1 /2 98 /0 19 02 01 / 97 / 9 10 1 /1 96 /0 9 06 1 /1 94 /0 19 02 01 / 93 / 9 10 1 /1 92 /0 9 06 1 /1 90 /0 19 02 01 / 89 / 9 10 1 /1 88 /0 19 06 01 / 86 / 9 02 1 /1 85 /0 9 10 1 /1 84 /0 19 06 01 / 82 / 9 02 1 /1 81 /0 9 10 1 /1 80 /0 19 06 01 / 78 / 9 02 1 /1 /0 10 Predicted values President Yes ARIMA prediction ARIMA versus OLS predicted values 80 70 60 50 40 30 20 Date 04 20 1 / 02 /0 20 02 01 / 01 / 0 10 1 /2 00 /0 20 06 01 / 98 / 9 02 1 /1 97 /0 19 10 01 / 96 / 9 06 1 /1 94 /0 19 02 01 / 93 / 9 10 1 /1 92 /0 19 06 01 / 90 / 9 02 1 /1 89 /0 19 10 01 / 88 / 9 06 1 /1 86 /0 19 02 01 / 85 / 9 10 1 /1 84 /0 19 06 01 / 82 / 19 02 01 / 81 / 9 10 1 /1 80 /0 19 06 01 / 78 / 9 02 1 /1 /0 10 OLS prediction Predicted versus actual values of presidential popularity from ARIMA and OLS models (Giscard) ARIMA predicted 70 60 50 40 30 Date 81 19 1/ /0 81 04 /19 1 /0 80 02 /19 1 0 /0 98 12 1 /1 /0 80 10 /19 1 /0 80 08 /19 1 /0 80 06 /19 1 /0 80 04 /19 1 9 /0 97 02 1 /1 /0 79 12 /19 1 /0 79 10 /19 1 /0 79 08 /19 1 /0 79 06 /19 1 9 /0 97 04 1 /1 /0 78 02 /19 1 /0 78 12 /19 1 /0 10 OLS predicted President Yes Predicted versus actual values of presidential popularity from ARIMA and OLS models (Mitterrand) ARIMA predicted 80 70 60 50 40 30 20 Date 94 19 1/ 4 /0 99 12 1 /1 93 /0 9 03 1 /1 92 /0 9 06 1 /1 91 /0 9 09 1 /1 91 /0 9 12 1 /1 90 /0 9 03 1 /1 89 /0 9 06 1 /1 88 /0 9 09 1 /1 88 /0 9 12 1 /1 87 /0 9 03 1 /1 86 /0 9 06 1 /1 85 /0 9 09 1 /1 85 /0 9 12 1 /1 84 /0 9 03 1 /1 83 /0 9 06 1 /1 82 /0 9 09 1 /1 82 /0 9 12 1 /1 81 /0 9 03 1 /1 /0 06 OLS predicted President Yes Predicted versus actual values of presidential popularity from ARIMA and OLS models (Chirac) ARIMA predicted 70 60 50 40 30 Date 04 20 1/ 3 /0 00 06 1 /2 03 /0 0 12 1 /2 02 /0 0 06 1 /2 02 /0 0 12 1 /2 01 /0 0 06 1 /2 01 /0 0 12 1 /2 00 /0 0 06 1 /2 00 /0 0 12 1 /2 99 /0 9 06 1 /1 99 /0 9 12 1 /1 98 /0 9 06 1 /1 98 /0 9 12 1 /1 97 /0 9 06 1 /1 97 /0 9 12 1 /1 96 /0 9 06 1 /1 96 /0 9 12 1 /1 95 /0 9 06 1 /1 95 /0 9 12 1 /1 /0 06 OLS predicted President Yes ARIMA and OLS predicted versus actual values, 1980 ARIMA predicted 60 58 56 54 52 50 48 Date 80 19 1/ /0 80 12 19 1/ /0 80 11 19 1/ /0 80 10 19 1/ /0 80 09 19 1/ /0 80 08 19 1/ /0 80 07 19 1/ /0 80 06 19 1/ /0 80 05 19 1/ /0 80 04 19 1/ /0 80 03 19 1/ /0 80 02 19 1/ /0 01 OLS predicted President Yes ARIMA and OLS predicted versus actual values, 1999 ARIMA prediction 64 62 60 58 56 54 52 50 48 Date 99 19 1/ /0 99 12 19 1/ /0 99 11 19 1/ /0 99 10 19 1/ /0 99 09 19 1/ /0 99 08 19 1/ /0 99 07 19 1/ /0 99 06 19 1/ /0 99 05 19 1/ /0 99 04 19 1/ /0 99 03 19 1/ /0 99 02 19 1/ /0 01 OLS prediction President Yes Correlation matrix of predicted values with actual value of presidential popularity Correlations ARIMA OLS predicted President Yes predicted 1 .912** .972** . .000 .000 280 280 277 .912** 1 .930** .000 . .000 280 311 279 .972** .930** 1 .000 .000 . 277 279 279 OLS predicted Pearson Correlation Sig. (2-tailed) N President Yes Pearson Correlation Sig. (2-tailed) N ARIMA predicted Pearson Correlation Sig. (2-tailed) N **. Correlation is significant at the 0.01 level (2-tailed). Why choose ARIMA? Linear forecast Random walk forecast ARIMA forecast Pitfalls of ARIMA  Complexities of model specification  Difficulties of interpretation  Sensitivity of data (e.g. missing values)