I plan to have 3 or 4 election prediction models by the first few days into the campaign and as we approach I’ll release them in draft form and continually make modifications as Judgement Day nears. Some will be pure economics, some will be pure opinion polls, some a mix and one will be out of left field.

Here’s the first draft of Election Prediction Model 1.

Note, all political variables come from Newspoll

For the Coalition:

GOVPRIMARY = dependent variable

GOVPRIMARY(-1) = the governments primary vote in the previous month

PMSAT = Satisfaction rating of the PM

IPDI-IPDI(-24) = interest payments to disposable income as a percentage minus the same measure 24 months previous.

INT = interest rates defined as the standard bank variable loan rate

GST is a dummy variable equaling 1 for each period when the GST has been in and zero otherwise.

RUDD = equals the “RUDD EFFECT” dummy variable equaling 1 when Rudd has been ALP leader and zero otherwise

CAMP = campaign dummy variable that measures 1 in the period the month before the election and zero at all other times.

The regression results tested over the Howard government period were:

 Dependent Variable: GOVPRIMARY Method: Least Squares Date: 05/25/07 Time: 17:41 Sample: 1996M03 2007M05 Included observations: 135 Variable Coefficient Std. Error t-Statistic Prob. C 12.94913 2.645991 4.893867 0.0000 GOVPRIMARY(-1) 0.239182 0.063741 3.752377 0.0003 PMSAT 0.269616 0.028716 9.389111 0.0000 IPDI-IPDI(-24) -0.493343 0.225858 -2.184312 0.0308 INT 1.126645 0.230097 4.896392 0.0000 GST -1.293712 0.439106 -2.946242 0.0038 RUDD -3.517838 0.836297 -4.206447 0.0000 CAMP 1.808813 0.917797 1.970820 0.0509 R-squared 0.730078 Mean dep var 43.110 Adjusted R-squared 0.715201 S.D. dep var 3.3607 S.E. of regression 1.793543 Akaike 4.0636 Sum squared resid 408.5332 Schwarz 4.2358 Log likelihood -266.2993 F-statistic 49.072 Durbin-Watson stat 2.130125 Prob(F-statistic) 0.000000

For the ALP Primary Vote:

The variables are:

Labor = The Dependent Variable

Labor(-1) = The ALP primary vote in the month previous.

OPSAT-OPDISAT = Oppositions satisfaction rating minus their dissatisfaction rating

IPDI-IPDI(-12) = the interest payments to disposable income percentage minus that same measure 12 months previous.

GOVPRIMARY = governments primary vote.

The regression results tested over the Howard government period were:

 Dependent Variable: LABOR Method: Least Squares Date: 05/25/07 Time: 17:10 Sample: 1996M03 2007M05 Included observations: 135 Variable Coefficient Std. Error t-Statistic Prob. C 33.59647 3.990479 8.419157 0.0000 LABOR(-1) 0.514651 0.059088 8.709883 0.0000 OPSAT-OPDISAT 0.050997 0.008650 5.895986 0.0000 IPDI-IPDI(-12) 0.335900 0.215453 1.559045 0.1214 GOVPRIMARY -0.335420 0.051636 -6.495840 0.0000 R-squared 0.747402 Mean dep var 39.986 Adjusted R-squared 0.739630 S.D. dep var 3.3356 S.E. of regression 1.702045 Akaike 3.9378 Sum squared resid 376.6042 Schwarz 4.0454 Log likelihood -260.8063 F-statistic 96.163 Durbin-Watson stat 1.915282 Prob(F-statistic) 0.0000

In order to forecast this into November, I first had to forecast into November any variables whose values these two primary vote models are reliant upon.

For “interest payments to disposable income” I used a conservative Holt-Winters method non-seasonal forecast to arrive at a final November forecast of 12.03 for this variable.

Interest rates I assumed to remain the same between here and November.Likewise PM satisfaction ratings I assumed would remain the same as they are now.Opposition satisfaction ratings I assumed would slowly decay starting from 68 today and ending at 60 by November, likewise the opposition dissatisfaction rating was assumed to slightly solidify starting from 18 today and increasing to 24 by November.These assumptions for satisfaction were based around previous movements of these elections variables in previous elections which I modelled separately.

Having ascertained the forecast values for my variables, I then forecast these two primary vote models into November using these values and the above regressions.

The results were:

Coalition on primary vote of 39.7

ALP primary vote of 45.6

I then assumed a preference distribution of 45/55 Coalition/ALP, which is slightly lower than usual, but chosen because of the higher ALP primary vote.

The final two party preferred forecasts became:

ALP 53.7 Coalition 46.3

Using Bryan of OzPolitics fames’ Election Calculator, the end results assuming a uniform swing would be the ALP picking up 86 seats and forming government.

To explain how the dummy variables work, the CAMP variable measures and accounts for the Coalitions better performance during election campaign in terms of improving their primary vote during the campaign.RUDD is a variable that measures and accounts for the observable boost that the Rudd leadership has on the ALPs primary vote.The GST variable measures and accounts for the consistent long term decay in the Coalitions primary vote since the implimentation of the GST.

Now for all you econometric type nit pickers out there, yes there is a small serial correlation in the error terms. This will be ARCHed out later closer to the election. Likewise, there are VAR opportunities, this model may well go up that route.

So what does the model forecast each month until November? 2 Party Preferred scale is on the right, Primary Vote scale is on the left:

(Visited 1 times, 1 visits today)