Budgets bounce like dead cats

I thought I’d use the newspoll results back to 1985 to see if the journo cliche of choice come May – “the budget bounce” was nothing more than a fantasy driven by people that have nothing better to say.

Surprise surpise, that’s the case.

So I also tested the excuse of choice for journo’s and governments alike when the polls dont react the way they apparently ought to – “its a delayed bounce, it hasnt happened yet, but it will happen”.

Surprise, surprise….. that’s twaddle too.

Using the series GOVPRIMARY (which is the newspoll measurement of the governments primary vote) as the dependent variable, I regressed it against 3 seasonal dummy variables S5, S6,and S7 representing the months of May, June and July respectively.If budgets actually do cause a bounce in the government vote in the polls, those seasonal dummy variables would be expected to have a positive coefficient.If those bounces happen as regularly as the commentariat believes, they should also be statistically significant i.e. the Prob value should be less than 0.1, preferably less than 0.05.

So let us see the results (a usual constant C was also added to the mix – just ignore it)

Again, using the quaint little Eviews to do the ugly math:

Dependent Variable: GOVPRIMARY
Method: Least Squares
Date: 05/17/07 Time: 17:05
Sample(adjusted): 1986:01 2007:05
Included observations: 257 after adjusting endpoints
Convergence achieved after 5 iterations
Variable	Coefficient	Std. Error	t-Statistic	Prob.
C	42.37000	0.688365	61.55166	0.0000
S5	-0.085694	0.502248	-0.170621	0.8647
S6	-0.686960	0.583794	-1.176716	0.2404
S7	-0.429398	0.511557	-0.839394	0.4020
AR(1)	0.774224	0.039983	19.36391	0.0000
R-squared	0.598817	Mean dep var		42.437
Adjusted R-squared	0.592449	S.D. dep var		3.8455
S.E. of regression	2.454982	Akaike		4.6533
Sum squared resid	1518.788	Schwarz		4.7224
Log likelihood	-592.9593	F-statistic		94.035
Durbin-Watson stat	2.216458	Prob(F-statistic)		0.0000

Lo and behold, all of the seasonal dummy variables are utter rubbish.They have no statistical significance and some of the coefficients are negative.Budgets dont bounce polls.

What about just over the Howard term of government.:

Dependent Variable: GOVPRIMARY
Method: Least Squares
Date: 05/17/07 Time: 10:43
Sample: 1996:03 2007:05
Included observations: 135
Convergence achieved after 3 iterations
Variable	Coefficient	Std. Error	t-Statistic	Prob.
C	43.11803	0.745668	57.82469	0.0000
S5	0.334513	0.688051	0.486174	0.6277
S6	-1.320608	0.805286	-1.639925	0.1034
S7	-0.321919	0.711838	-0.452237	0.6519
AR(1)	0.713235	0.063750	11.18798	0.0000
R-squared	0.494724	Mean dep var		43.110
Adjusted R-squared	0.479177	S.D. dep var		3.3607
S.E. of regression	2.425422	Akaike		4.64622
Sum squared resid	764.7476	Schwarz		4.7538
Log likelihood	-308.6200	F-statistic		31.821
Durbin-Watson stat	2.030240	Prob(F-statistic)		0.0000

Again – utter twaddle.

Budgets dont give governments a bounce in the polls.It is not in any way an expected voter behaviour

UPDATE :GriffithGuy raised an interesting question.Do budgets bounce in election years when the pork is flying?

Lets test the theory again, but this time using dummy variables representing only budgets in election years.First for all governments since 1986:

Dependent Variable: GOVPRIMARY
Method: Least Squares
Date: 05/18/07 Time: 01:17
Sample(adjusted): 1986:01 2007:05
Included observations: 257 after adjusting endpoints
Convergence achieved after 6 iterations
Variable	Coefficient	Std. Error	t-Statistic	Prob.
C	42.27549	0.684618	61.75049	0.0000
S5E	0.466934	0.832840	0.560653	0.5755
S6E	-0.653364	0.995129	-0.656562	0.5121
S7E	-0.318264	0.880077	-0.361632	0.7179
AR(1)	0.774742	0.040092	19.32387	0.0000
R-squared	0.598543	Mean dep var		42.437
Adjusted R-squared	0.592170	S.D. dep var		3.8455
S.E. of regression	2.455822	Akaike		4.6540
Sum squared resid	1519.827	Schwarz		4.7231
Log likelihood	-593.0472	F-statistic		93.928
Durbin-Watson stat	2.228653	Prob(F-statistic)		0.0000

Again, no budget bounce as the coefficients are small and mostly negative, and they are statistically insignificant.
But what about Howard from 1996-2007?

Dependent Variable: GOVPRIMARY
Method: Least Squares
Date: 05/18/07 Time: 00:59
Sample: 1996:03 2007:05
Included observations: 135
Convergence achieved after 5 iterations
Variable	Coefficient	Std. Error	t-Statistic	Prob.
C	43.04898	0.755544	56.97745	0.0000
S5E	1.098800	1.051503	1.044980	0.2980
S6E	-2.281707	1.285617	-1.774795	0.0783
S7E	-0.939306	1.151982	-0.815383	0.4163
AR(1)	0.723267	0.063517	11.38701	0.0000
R-squared	0.504229	Mean dep var		43.110
Adjusted R-squared	0.488975	S.D. dep var		3.3607
S.E. of regression	2.402501	Akaike		4.6272
Sum squared resid	750.3613	Schwarz		4.7348
Log likelihood	-307.3381	F-statistic		33.054
Durbin-Watson stat	2.065642	Prob(F-statistic)		0.0000

Again no bounce as the coefficients are insignificant, but at the 10% level of significance Howard actually loses over 2% off his primary vote in the June following an election year budget.

Ouch.