## I Love the Smell of Interest Rates in the Morning

Posted by Possum Comitatus on August 22, 2007

Well not really, unless I was a self-funded retiree – but before we head down that route, let’s fold the latest Newspoll into our monthly Newspoll average and see what’s been going on. The Libs and the ALP primary vote can be read from the left hand side, the Nats can be read from the right.

A great load of absolutely nothing happened. The absence of movement is becoming more important than movement itself. It’s better visualised with the following two graphs:

February 2006 was the last time that the Coalition was ahead on TPP terms – that’s 18 months ago.

Questions of trends arise here. Has there been a trend away from the ALP since May 2007? Has the TPP started to asymptote, as in flattened out at a level of solid political support?

It depends on the way you analyse it. It if you treat each observation as completely independent, then there has been a trend away since May. If you treat the observations as an actual time series where each observation isn’t completely independent, but is a partial product of observations that came before it – then the answer is no. The movement since December cannot yet be separated from polling noise, margin of error and the usual small meandering around that poll estimations produce over time if you treat the polling series for what they are – time series.

If I could point out the obvious here, the longer the status quo remains, the more difficult it becomes for the Coalition to get the numbers they need to retain government. At the moment, to even have a remote chance of winning the election, the Coalition needs to claw back 5 points of primary vote while the ALP needs to lose 5 – yet that wouldn’t guarantee a Coalition victory, just that it would be possible as long as they picked up a slightly higher than usual preference flow. A movement in the primary vote of this size over this time frame has been done once by Howard – between June and September 2001 when the Tampa and September 11 were happening. The Coalition gained 6.5 points and the ALP lost 5 points from their respective primaries back then

The only other time that comes close is in the period of August to October 2002 where the Coalition gained five and the ALP lost 5.Yet that is more likely than not a margin of error issue as there was only one Newspoll in October 2002 and that gave the Coalition a primary of 47. They jumped to 47 from a monthly average of 41.5 in September and in November they dropped back to 45.5 then to 43 in December of that year.

So Howard has to do what he has only done once before – but without Tampa and S/11 to help him.

But getting back to interest rates, how do interest rate rises affect the vote estimations of the two major parties?

Since the Newspoll public database started in December 1985, the coalition primary vote has generally benefited by a fraction of a point following an interest rate rise and suffered by a fraction of a point following an interest rate fall at about the 5% significance level. Likewise, over just the period of the Howard government, the Coalition primary vote goes up with rate rises and goes down with rate cuts by a small fraction of a percent at the 10% level of significance.

But since 2003, things have changed.

If we regress the government primary vote on a constant and the banks standard variable interest rate using the Newspoll data we get:

Dependent Variable: Coalition Primary: Sample (adjusted): 2003M01 2007M08: Included observations: 55 after adjustments

Variable |
Coefficient |
Std. Error |
t-Statistic |
Prob. |

C |
69.08248 |
4.247783 |
16.26318 |
0.0000 |

INT |
-3.674950 |
0.584621 |
-6.286039 |
0.0000 |

R-squared = 0.43

This suggests that a standard 0.25 percent interest rate increase takes nearly 1% off the governments primary vote. That drops to about to about 0.5% if we take into account the autoregressive nature of the polling series (where we add the value of the Coalitions primary vote in the previous month to the above regression equation)

If we do the same using Morgan data we get:

Dependent Variable: Coalition Primary:Sample: 2003M01 2007M08:Included observations: 56

Variable |
Coefficient |
Std. Error |
t-Statistic |
Prob. |

C |
78.19676 |
3.524834 |
22.18452 |
0.0000 |

INT |
-5.380761 |
0.484139 |
-11.11408 |
0.0000 |

R-squared = 0.70

Which suggests a larger effect, with a 0.25% increase in rates taking 1.3% off the governments primary vote.That too drops to around 0.5% if we take into account the autoregressive nature of the polling series.

To give you an idea how that looks, the graphical results of the above equation using Morgan is:

So we can say that if interest rates have a causal effect (which Granger causality tests support at the 2% level of significance), according to the two most regularly published polls, a usual 0.25 percent rise in interest rates reduces the Coalition primary vote by between 0.5 and 1%.

So the question becomes – when does the effect happen?

The useful thing about finding out when it happens is that it also provides us with more information about how the interest rate rises spike the vote of the two major parties. We already know that on average a 25 basis points rise in interest rates lead to an average decrease in the Coalitions primary vote of between 0.5 and 1%.But that “average” wont be distributed evenly, it will spike before settling down to give us that average effect level.

To find out the timing and spike effects, I’ve created a dummy variable called “dummyraterise” which has a value of 1 in the first poll taken after the interest rate rise and a value of zero all other times.

We then regress the estimated vote on a constant C, the estimated vote in the previous poll (to account for the autoregressive inertial effects of the voting series) and on this dummy variable. We’ll use the same time period here starting in Jan 2003and using every poll through to the current poll.

First off, the ALP primary vote using Newspoll:

Dependent Variable: ALP PRIMARY:Sample (adjusted): 315 428:Included observations: 114 after adjustments

Variable |
Coefficient |
Std. Error |
t-Statistic |
Prob. |

C |
6.721322 |
2.257646 |
2.977138 |
0.0036 |

ALPPRIM(-1) |
0.831032 |
0.055986 |
14.84363 |
0.0000 |

DUMMYRATERISE(-2) |
2.671309 |
1.029682 |
2.594306 |
0.0108 |

R-sq = 0.66

After checking all the lagged periods of the rate rise for significance, I found throughout these regressions that the only statistically significant period for the rate rise effect kicking in was 3 polls after the rate rise had occurred (the first poll after the rate rise plus 2 lagged polls later… hence the (-2) after DUMMYRATERISE)).

This suggests that a rate rise increases the ALP primary vote by about 2.5 points, 3 Newspolls after the rate rise occurs.

Yet when we try the Coalition with Newspoll, there is no significant dummy variable effects at any lag. This suggests that the drop in the Coalition vote that occurs with interest rate rises happens slowly over time on the basis of the Newspoll.

So where does the ALP spike in the vote that occurs after a rate rise come from according to Newspoll?

It comes from the “Others”…. the minor parties.

Dependent Variable: OTHERS PRIMARY: Sample (adjusted): 315 428: Included observations: 114 after adjustments

Variable |
Coefficient |
Std. Error |
t-Statistic |
Prob. |

C |
6.397435 |
1.317989 |
4.853938 |
0.0000 |

OTHERSPRIM(-1) |
0.637503 |
0.074541 |
8.552395 |
0.0000 |

DUMMYRATERISE(-2) |
-1.903743 |
0.892724 |
-2.132510 |
0.0352 |

R-sq=0.40

According to Newspoll, when an interest rate rise occurs, 3 polls after it is announced the ALP primary gets an average boost of 2.5 points, about 2 points of which comes from the minor parties. The other half a point comes from the Libs or uncommitted voters.

How does the Morgan reading stack up:

Dependent Variable: ALP PRIMARY :Sample (adjusted): 186 320 :Included observations: 136 after adjustments

Variable |
Coefficient |
Std. Error |
t-Statistic |
Prob. |

C |
8.424555 |
2.303184 |
3.657787 |
0.0004 |

ALPPRIM(-1) |
0.798993 |
0.054423 |
14.68110 |
0.0000 |

DUMMYRATERISE(-2) |
2.082629 |
1.104975 |
1.884774 |
0.0617 |

R-Sq=0.62

Dependent Variable: COALITIONPRIM :Sample (adjusted): 185 320 : Included observations: 136 after adjustments

Variable |
Coefficient |
Std. Error |
t-Statistic |
Prob. |

C |
9.866994 |
2.353931 |
4.191710 |
0.0001 |

COALITIONPRIM(-1) |
0.765045 |
0.055896 |
13.68692 |
0.0000 |

DUMMYRATERISE(-2) |
-1.778497 |
0.996178 |
-1.785320 |
0.0765 |

R-sq=0.58

The “others” have no significant effects. There are two things interesting here. Over the period tested, there are more Morgans than Newspolls, yet the both suggest that it takes 3 polls from the moment interest rates rise for that effect to turn up in their respective polling. The other thing of note is that whereas Newspoll registers a spike in the ALP primary vote when interest rates rise and suggests that it comes at the expense of the “others” vote, Morgan suggests that a slightly smaller spike flows to the ALP but comes at the expense of the Coalition primary vote.

When we test the ALP TPP vote against interest rate rises, we get a 1.6 percent spike in the ALP vote, 3 polls after the interest rate rise occurred at a 10% level of significance.

So if history holds, there should be a small increase in the ALP vote: at the beginning of September for Morgan and at the end of September for Newspoll. Although the suggested spike would be smaller than the margin of error in either poll.

Also something to keep an eye on: Peter Brent of Mumble fame is doing something rather interesting over **HERE** and it will be well worth watching.

**UPDATE **

Theres a new Morgan phone poll out today (the 23 August)

ALP 60/40 on TPP, 51/36 on primaries.

It was taken over the 21st and 22nd.

Right on queue, the third Morgan poll after the interest rate rise showed an increase in the ALP vote… although the poll came earlier than expected.

I’d like to think it was an accurate prediction 😉

## Stunkrat said

First pedant off the block – Nats from the right, ALP & Libs from the left.

## Possum Comitatus said

Doh!

## steve said

If the betting markets are correct then Election Day 2007 is likely to be either 10 November or 24 November. This would allow for either 6 or 7 more fortnightly polls of voters. As noted above if the gap of 10% or so between the TPP of the two main groups remains then there is little doubt of the final outcome (a uniform 4.1% swing delivers 16 seats).

The 4 to 5 weeks of the actual election campaign (as opposed to what is now occuring) has in the past moved voting trends by between 2 to 4% but not 10%.

We see that the promises and dollars have started to flow. I suspect that the rate funds will flow from announced budget reserves will increase to an even faster level than they ever have before.

It will be most interesting to see the effect on voters of promises of future funding for pet electorate projects versus the impact of increased interest rates. Increased future spending does have an impact on future inflation (RBA have advised they will raise rates if they identify a need to quell inflation) so it will come down to how this process is managed. One more interest rate increase would move the voting trends to End Game but what other option does the Government have.

## canberra boy said

Possum, I e-mailed you the other day to enquire about the effect of the recent interest rate rise on your election prediction model. That model was based on evidence that interest rate rises (as economic ‘bad news’) benefited the incumbent government. You’ve now suggested in this post that since 2003 interest rate rises have hurt the polling support for the Coalition. So do you want to revisit the model? And how does the latest RBA quarterly interest rates to disposable income figure affect the outcome. I had been expecting a tension between a voter flight to the devil-you-know versus the voter reaction to the squeeze on disposable income.

## Mate said

Thank you my friend.

Excellent, as always

## Stig said

Good stuff as always.

I’m not sure how observable any effect from this rate rise will be in the polls you note though. We can anticipate a few other large-scale things potentially affecting voting intentions; along the lines of APEC, the possible formal election campaign, and shocking revelations of Rudd having a couple of beers over dinner at his daughter’s wedding. So, this one may have to go into the “Unprovable” file as far as predictive value goes.

However, it is useful to have a describable cause & effect based on data when it comes to interest rates, to contrast with some of the more hallucinatory opinions out there in both the MSM and the blogosphere. I’m wondering whether some sort of introductory stats should be taught in journalism schools…

## Possum Comitatus said

Canberra boy – those bloody interest rates are giving me grief with the models.Over the very long term the interest rate/primary vote relationship goes one way, over the medium term it goes the other way.I’m going to have to play around with it and fix it up to account for that quirk.

I agree with you over the tension, if only we had more elections the data might be able to tell us something.

Stig – I’m not sure either.Something moved the last Morgan beyond the MoE, whether that was Costello’s drunken ego, the rate rise or a combination or something else entirely is impossible to tell.The problem with these things is that they aint very predictive.Alot gets written about reactions to interest rates – including by me in the last Morgan post.It’s nice to be able to bed down some statistically significant reality.

## lurker said

The next Morgan telephone is out with 60/40. By my reckoning that’s the third Morgan poll after the rate rise – so your predictions of it taking 3 polls to register a rate rise are held up by this poll.

## Possum Comitatus said

Thanks lurker. Who are you – The Flash?

You managed to point this out before the traffic links came in.

That’s some serious surfing speed mate!😉

## Possum Comitatus said

Actually Lurker you are right – this is the third Morgan poll after the interest rate rise and up she went.

Righty’O, where are my Tarot cards… I want next weeks lotto numbers😉

## lurker said

Twasn’t me. Was someone on the Oz election site!

## Possum Comitatus said

S u r e Lurker,

That’s what all the guys that wear their budgie smugglers on the outside of their unitard say😉

## Leopold said

Can I ask… in that first graph, are the points the monthly Newspoll averages or are they calculated some other way? The Labor primaries just look a bit odd to me.

Second, are you really comfortable with using such data for regression purposes? Some of the observations would have 4 polls in them (May 2007) some just 1 (June 2007) which raises questions about their consistency. Wouldn’t a moving average over 3 or 5 periods be better? I know it tends to lag the data, but the movements could be tracked without the arbitrariness that month-to-month measurement causes.

## Stig said

OK, well I guess this model did hold up in a predictive way after all. So well done on that!

I’m still having trouble with the numbers on this Morgan poll – both the voting intention and particularly Howard’s personal ratings have moved more than I would have thought was credible. However, if it is driven by an issue hitting home with voters then it gets more believable. The media fluff over the last week wouldn’t explain this movement, but the delayed effect from an interest rate rise would. So maybe it is real.

Back on the huge change in Howard’s personal ratings – that is spectacular. I’d interpret it as voters feeling they have been lied to, in person, by Howard at the last campaign about interest rates, and that his greasy wriggling about with weasel words on the subject isn’t cutting it with the electorate. I don’t expect to read that interpretation in the GG though, it’s just another hallucinatory opinion out in blog land.

## Possum Comitatus said

Hi Leo – thanks for that.I’ve wondered if that graph was right a couple of times and thought… nah – it’s gotta be right.

But it was wrong.The June entry was wrong.I used a new version of my eviews software when I was doing it and I might have stuffed some manipulation up.The data is still fine in my database (that’s where the replacement came from), so it must have been something I did to it in the program itself.

Million thanks.

I’ve got every poll so I use which ever type of data amalgamation suits the analysis.For the longer term trends, months is a handy unit.Monthly averages will have some months being more stable than others simply as a result of there being more polls in that month.But all that really means in practice is that some months will have the issue of a full blown 3% MoE while others will have some, unknown amount less than that.If I want to use a consistent time measurement, I have to sacrifice a bit of stability.If I want stability I would sacrifice some amount of time consistency.That’s the trade off.

Moving averages destroy information in the data as there is still alot of information in the volatility – its just harder to get out.And since alot of the things I look at have quite sharp movements, a moving average would cloud the enormity of some of those movements.

In an ideal world, Newspoll would be really nice and give exactly 2 polls a month, no more, no less😉

But over longer time frames, the MoE issues cancel each other out from month to month, or simply act against the regression.So I dont think it gives me bad results with my regressions, but gives me results that are probably less strong, and the variables less statistically significant than is the true case.e.g I might have some variable with a p-value of 0.07, even though the real, actual significance is at the 5% level rather than the 7% I end up with.

But I dont like moving averages of any description with regressions.It kills info – I’m of the school of it being better to more info than less, even if it becomes a bit of a pain to pull it out properly.

## Leopold said

I guess I should note that I do record moving averages (both of individual polls and an aggregation of non-Morgan ones) and in relation to the point about whether or not we’re seeing movement:

5-period Newspoll average peaked for Labor at 49.8-36; now says 47-39.2.

3-period Newspoll average peaked for Labor at 50.3-35.7; now says 47-39.3.

Using the aggregated figures gives slightly higher peaks and a slightly worse position now.

Obviously this is all arguable, but even if it’s not 95% certain, I reckon we can look at 2 standard deviations and say there’s a 68% probability (if my memory serves me rightly) of significant movement.😉

And I’m not enthused by the idea of using MAs for regression either, but the monthly averages are depressingly arbitrary.

## lurker said

If the monthly averages seem arbitrary i.e. more polls in one month than the next, couldn’t you use some type of weighting to the data to reflect the different precision in each month? If you had the numbers surveyed in each poll you could work out the precision, more work though I know for so many polls.

## canberra boy said

Hey, be careful Possum – the “third Morgan poll after the interest rate rise showed an increase in the ALP vote” but it was 1.5% and the margin of error in the poll would be around 4% on a sample size of 633 this time and 589 last time. Not a statistically significant change! Still, one’s little, pounding public servant heart is heartened…

## Possum Comitatus said

Leo, the trends are all about how you treat your data and how you define “trend”.If you define trend as some literal interpretation of a run of an arbitrary number of observations of (on average) decreasing value – you’ll have a trend.But that descriptive ,literal interpretation can be applied to all sorts of short term sequences embedded in a longer term series that doesnt actually explain what is happening to the data.

If you define a trend as an explanatory concept that provides value in describing the underlying behaviour of the data – that gives you 2 possibilities.

1.If you treat each observation as completely independent, the evidence will suggest a trend back.

2.If you treat each observation as being merely a point in a time series which exhibits auto and partial autocorrelation and where there is a significant stochastic component – then the available evidence is insufficient to be able to claim a ‘trend’ since march.There is not enough evidence to state that a ‘trend’ since March exists… in fact, that type of short term sloping movement is to be expected throughout the series.

Monthly averages are depressingly arbitrary – but they are handy if a time consistent unit can be used to help explain how some process changed over a period of time.One of the benefits of using them is that their primary problem is one of underestimating effects and relationships and their significance.It’s like having an extra layer of conservatism applied to the process of determining results.

Lurker – I could build a seperate series that had two values, the variable that is being measured (say the Coalition primary vote) and an approximate measure of its uncertainty by month.That would then allow for some types of analysis that use the primary vote while accounting for the uncertainty of each observation.But that would remove a whole school of econometric approaches from the armory and wouldnt greatly improve the ability to find really significant relationships.And it would be alot of time to build, and some of the polling data from years past no longer has the information on sample size and whatnot available.

CB, yeah I know.Unfortunately we’ll never be able to truly tell whether any poll had an interest rate effect for the very reason you say (the effect is less than the MoE), just that over history we know that one does exist and roughly the time frame it works on.

That’s why I put in that smiley when I said “I’d like to think it was an accurate prediction”.The real beauty here is that unless there was an increase in the support for the Coalition by 2 points or more, I could still claim that interest rates had an effect but they were just clouded by the MoE.Then again, if there was a movement greater than 2 points to the coalition I could have claimed that it was a one in twenty “rogue poll”!😉

## Possum Comitatus said

Hi stig – the Morgan is interesting, and even taking into account the smallish sample, some of those movements are still significant and large.

The big change in Howards satisfaction issues is compatible with a higher ALP vote (assuming a real increase was actually there for a second)if interest rates where the driver of the lift in support for the ALP.Its also compatible with an anti-wowserism and disdain for muckracking streak in the electorate.We’ve seen with the Burke affair and the like that the electorate doesnt appreciate it, and from the survey responses to strippergate, it seems they thought that was twaddle as as well.It might just be a combination of all three blowing back onto the government.

I’ve been thinking lately that Morgan seems to be a poll that embodies first impressions in its results.When a news story comes through that shifts votes in all the polls, it seems to do so first in Morgan and does so to a larger extent in Morgan compared to other polls.

I find myself starting to look at the polls through a prism of Morgan being the best at measuring first impressions, Newspoll being the best at measuring how those first impressions translate through to serious voter change and ACN being the best at showing a long term, conservative estimation of how that voter change becomes solid support.

I still cant get my head around how Galaxy fits into the equation though.I know Galaxy are great pollsters and that Briggs is exceptional at what he does, but I still cant quite fit them into the polling web neatly.They’d probably be proud of that actually😉

## Stig said

I quite like this interpretation of the different polls this way. Maybe there is some way to test it – seeing if Morgan movements lead Newspoll movements, etc? Obviously not all changes in Morgan would lead to longer-term trends, but you’ve already identified several drivers of long term effect on the polls over the last 10 years – they could be examined to see which polls move first, and how long that movement is observable in each of the polls. Just thinking…

Also, check out Bryan’s site for his thoughts on Morgan, and differentiation of his phone and face-to-face polling. He’s got the F2F showing similar trends to the other majors, but the phone poll bouncing around with no long term correlation (I think – I’ll check it again later). That would fit with your idea about it being a sensitive indicator of first impressions, yeah?

## Pi said

GST Effect in 2000/2001? I thought the GST was in 1998.

## Possum Comitatus said

Pi – the GST was implemented on July 1st 2000.It was taken to the 1998 election as a policy, but didnt actually come into operation until 2000.

Stig – I’m playing around for a way to test it.The problem is that over any given time period, the two polls move together anyway.That creates the problem that in the general case, Newspoll will be a leading indicator of Morgan, and Morgan will be a leading indicator of Newspoll.

I can use the first differences, but that gets distorted by MoE issues.I’ve got a couple of ideas, and if any of them work I’ll whack them up.It seems like such a simple problem but unfortunately its actually not at all because of the uncertain nature of the polling data and the fact that polls from different pollsters in large part move together.

## Psychiatria said

You really make it seem so easy with your presentation yet I find this matter to be seriously something that I believe I’d never understand. It seems too difficult and highly broad for me. I’m looking forward for your next write-up.