The Discrepancy

[fellow]

 

[Armus]

Pretty sure Charlie and Housing Stig don't know what they're talking about.  I plan to go back and compare September 2022 to September 2023 using LR data in Q1 next year and the difference will not be so significant.  This is really a non-issue.  They should have at least reviewed the LR methodology information online and given the LR the right of reply.

Edit - the reason I say they don't know what they're talking about is when Stig refers to confidence levels but I think he is misusing / misunderstanding that terminology.

The bit they may have a valid point for is mix-adjusting / seasonal adjusting not being accurate / useful in a downturn.

Edited November 3, 2023 by Armus

[nero120]

1 minute ago, Armus said:

Pretty sure Charlie and Housing Stig don't know what they're talking about.  I plan to go back and compare September 2022 to September 2023 using LR data in Q1 next year and the difference will not be so significant.  This is really a non-issue.  They should have at least reviewed the LR methodology information online and given the LR the right of reply.

Hi newbie, which estate agent do you work for then? 😂

[Smith]

Goes to show the importance of looking at the raw data, rather than being swayed by the commentary. In this case, the raw Land Registry figures show that sold prices have dropped by 12% in a year (19% for flats and terraced houses).

[Armus]

1 minute ago, nero120 said:

Hi newbie, which estate agent do you work for then? 😂

HPC forum standard / classic reply.  Been lurking / posting sporadically on HPC for 10+ years (maybe 15).  

Not denying there is a potential crash on just saying comparing 78k transactions against 6k transactions and not allowing all the registrations for that month to filter into the data is deeply flawed. 

[nero120]

Just now, Armus said:

HPC forum standard / classic reply.  Been lurking / posting sporadically on HPC for 10+ years (maybe 15).  

Not denying there is a potential crash on just saying comparing 78k transactions against 6k transactions and not allowing all the registrations for that month to filter into the data is deeply flawed. 

Did you watch the video?

[Armus]

Just now, nero120 said:

Did you watch the video?

Yes, all of it.

[14stFlyer]

1 minute ago, Armus said:

Yes, all of it.

Well done!  Better than most of us that reply to this sort of you tube video thread!

[nero120]

7 minutes ago, Armus said:

Yes, all of it.

Well they state that, statistically speaking, using 10% of a data set you get a 95% chance that the result will be ±1.5% using the full set. Do you have a particular issue with that general rule?

And given that the ONS use the same amount of results to opine on "house prices", why not also criticise the ONS? All Charlie and Stig are doing is removing the ONS model and looking at the raw data, why is that so bad?

The fact you say they "don't know what they are talking about" betrays a large amount of bias against them, speaking to incentive...

Edited November 3, 2023 by nero120

[fellow]

8 minutes ago, Armus said:

Pretty sure Charlie and Housing Stig don't know what they're talking about.  I plan to go back and compare September 2022 to September 2023 using LR data in Q1 next year and the difference will not be so significant.  This is really a non-issue.  They should have at least reviewed the LR methodology information online and given the LR the right of reply.

Edit - the reason I say they don't know what they're talking about is when Stig refers to confidence levels but I think he is misusing / misunderstanding that terminology.

The bit they may have a valid point for is mix-adjusting / seasonal adjusting not being accurate / useful in a downturn.

The correlation between the average mortgage approval price data and the raw LR data was undeniable (almost a one to one correlation), and clearly shows house prices will fall significantly over the next three months (unless people have managed to pluck massive deposits out of thin air).

[14stFlyer]

Just now, nero120 said:

Well they state that, statistically speaking, using 10% of a data set you get a 95% chance that the result will be ±1.5% using the full set. Do you have a particular issue with that general rule?

Works if the sample is random or “representative” of the full population.  In this case, we are looking at the first 10% of transactions to be processed. Not random.  This said, they could still be representative.  The point I am making is that this is not an appropriate example on which to impose your 95% confidence interval. 

[nero120]

1 minute ago, 14stFlyer said:

Works if the sample is random or “representative” of the full population.  In this case, we are looking at the first 10% of transactions to be processed. Not random.  This said, they could still be representative.  The point I am making is that this is not an appropriate example on which to impose your 95% confidence interval. 

Again, ok for the ONS to make a report on general "house prices" but not ok for Charlie/Stig to opine on?

[Smith]

15 minutes ago, Armus said:

Edit - the reason I say they don't know what they're talking about is when Stig refers to confidence levels but I think he is misusing / misunderstanding that terminology.

I also noticed that Alex's use of the term confidence intervals was a bit dubious. But the point he was making is nonetheless clear: that the initial 10% or so of data points posted by Land Registry for a given month have proved to be representative of the whole population, plus or minus a few percentage points. Do we know if this claim is accurate? If so, then I think they're analysis of the means and medians is fair.

[mynamehere]

It was only a month ago Charlie was citing the 10% sample as a reason not to trust land registry reports until all he data comes in so don’t judge him till May 24…. Now he is busily googling confidence levels so he can conjure up some new angle based on raw averages?

I would have thought in a period of mid volatility then the raw average becomes less reliable, not more? Did I miss his point?

mortgage approvals indices presumably don’t have the issue? 

Land registry seems to be about in sync with all the other indices once you adjust for lag

The indices that publish regional data seem aligned with what I’m seeing on the ground  

I don’t doubt that flats are down a lot but even his favourite raw data shows overall just a 6% drop from peak, which most people are already expecting  when the 2022 peak finally rolls into the land registry data?

And after all of it he concedes that the market has real wage growth over the period so it’s not clear how this fits into his arguments! He seems rather happy to move on to other subjects whenever Alex brings it up …0:

 

 

 

[nero120]

20 minutes ago, Smith said:

I also noticed that Alex's use of the term confidence intervals was a bit dubious. But the point he was making is nonetheless clear: that the initial 10% or so of data points posted by Land Registry for a given month have proved to be representative of the whole population, plus or minus a few percentage points. Do we know if this claim is accurate? If so, then I think they're analysis of the means and medians is fair.

Wait so the argument is not that they are incorrect about a >10% fall from peak, but that it's only the first 10% of transactions and therefore not representative of the full set?

In that case, the full set could be at even greater levels of annual falls?

So... what's the issue here again?

[shlomo]

35 minutes ago, 14stFlyer said:

Well done!  Better than most of us that reply to this sort of you tube video thread!

Can I reply without viewing the video 

[14stFlyer]

1 minute ago, shlomo said:

Can I reply without viewing the video 

It’s a free world.  Well, some of it.  

[scottbeard]

37 minutes ago, 14stFlyer said:

Works if the sample is random or “representative” of the full population.  In this case, we are looking at the first 10% of transactions to be processed. Not random.  This said, they could still be representative.  The point I am making is that this is not an appropriate example on which to impose your 95% confidence interval. 

+1

There are a lot of armchair statisticians on this internet, and many on this forum.  The analysis of excess deaths threads have similar.

[msi]

16 minutes ago, scottbeard said:

+1

There are a lot of armchair statisticians on this internet, and many on this forum.  The analysis of excess deaths threads have similar.

Don't forget the armchair economists, generals, virologists, and, of course, our armchair politicians

[Armus]

1 hour ago, 14stFlyer said:

Works if the sample is random or “representative” of the full population.  In this case, we are looking at the first 10% of transactions to be processed. Not random.  This said, they could still be representative.  The point I am making is that this is not an appropriate example on which to impose your 95% confidence interval. 

@nero120 yes, that’s my issue. Confidence level is WITHIN a sample. You can’t say sample 2 (September 23) is 10% of the size of sample 1 (September 22) and therefore gives a 95% confidence level. They don’t have the full population for September 2023. The current transactions might be representative, they might not be.
 

I believe you also have to calculate a sample size to achieve a required confidence level, it’s not a blanket number.

Stig asserting something doesn’t make it true.

I’m sure there may be issues with the ONS model (I already said that) but they have a methodology to try to make the data comparable rather than just using an incomplete data set.

@fellow - I’m not saying prices won’t drop. Look at my posting history and my views on where house prices should be are pretty clear.

Edited November 3, 2023 by Armus

[nero120]

19 minutes ago, Armus said:

@nero120 yes, that’s my issue. Confidence level is WITHIN a sample. You can’t say sample 2 (September 23) is 10% of the size of sample 1 (September 22) and therefore gives a 95% confidence level. They don’t have the full population for September 2023. The current transactions might be representative, they might not be.
 

I believe you also have to calculate a sample size to achieve a required confidence level, it’s not a blanket number.

Stig asserting something doesn’t make it true.

I’m sure there may be issues with the ONS model (I already said that) but they have a methodology to try to make the data comparable rather than just using an incomplete data set.

@fellow - I’m not saying prices won’t drop. Look at my posting history and my views on where house prices should be are pretty clear.

Fair enough.

So in conclusion, we all categorically agree that the 10% of transactions in the latest data are down >10% from peak.

[Smith]

20 minutes ago, Armus said:

@nero120 yes, that’s my issue. Confidence level is WITHIN a sample. You can’t say sample 2 (September 23) is 10% of the size of sample 1 (September 22) and therefore gives a 95% confidence level. They don’t have the full population for September 2023. The current transactions might be representative, they might not be.
 

I believe you also have to calculate a sample size to achieve a required confidence level, it’s not a blanket number.

Stig asserting something doesn’t make it true.

I’m sure there may be issues with the ONS model (I already said that) but they have a methodology to try to make the data comparable rather than just using an incomplete data set.

@fellow - I’m not saying prices won’t drop. Look at my posting history and my views on where house prices should be are pretty clear.

Confidence intervals are used to work out a likely range for the population mean (in this case, all September transactions) from the sample mean (in this case, the September transactions we know about so far) and standard deviation. So there's nothing wrong with this approach in principle. The issue is whether it's accurate to say this gives a range of only 2% in either direction. Rather than calculate this mathematically, it would be easier in this case to look at initial vs. final data from previous months and see how much the mean typically changes. If the final mean is never more than a few thousand different from the initial mean then their analysis is fair.

[mynamehere]

It's a bit of a side track, the confidence interval isn't really the core of his claim. Though I'm sure if you were to look back at his output it wouldn't be hard to find him complaining about small sample sizes when it suits.

It's a simple complaint that mix adjustment works differently on the way up than on the way down.

Nationwide publish a very accessible explainer here of why mix adjustment is required,

https://www.nationwide.co.uk/-/assets/nationwidecouk/documents/about/house-price-index/nationwide-hpi-methodology.pdf?rev=eb1695a5f9df46fa815b5810371c4785

They mention that typically they refine the mix forumula every two years ago, but obviously things changed a bit more quickly this year. 

I raised the issue on a land registry thread back in Q1 and showed my own research into the raw data showing that revisions had spiked from typical 0.5% annualised to 5% annualised and were rising,

However land registry made changes to their algorithm shortly after and the revisions settled back to historical norms

Quite why charlie thinks LR are incapable of refining their index during a downturn remains a mystery, they presumably have a large team working on it

As for the question why the mortgage approval index which is mix adjusted is not showing the same falls as his mortgage derived raw data, Charlie again doesn't have an answer other than 'they must be doing it wrong'

This is really basic stuff and I still beleive charlie is intentionally playing dumb about this stuff. As this thread shows, it certainly gets views and clicks.

Perhaps he is unconsicously biased towards opinions that get him clicks

 

Edited November 3, 2023 by mynamehere

[Sackboii]

3 hours ago, Armus said:

Not denying there is a potential crash on just saying comparing 78k transactions against 6k transactions and not allowing all the registrations for that month to filter into the data is deeply flawed. 

And yet the same situation of a massively reduced amount of transactions is oft used to as an indicator that prices are rising.

There's only one way to know which way things are going other than looking at leading indicators and reduced transaction numbers, and that's to wait for all of the data, which in many cases is *at least* six months behind now.

We can't have a situation where 'initial data' is used for the VI 'prices are rising' pump and then the same poo-poo the same data when prices are falling, or argue that the initial data is 'as good as it needs to be and is pretty accurate' when in fact many months later it gets revised with more up-to-date data which changes the narrative.

In short, agreeing with you. ☺️

[sell2rent]

95% CI of +-1.5% would involve the total number as well as a percentage of the sample, it would also rely as people have already said on the first 10% not being biased (if the next tranche shows similar the precision would increase), and also relies on all of it being a normal distribution which it probably isn't. It might not be invalid, but it is prematurely drawing a conclusion. It may be making an interesting idea to watch to firm up later. Not a statistician, but had to calculate 95% confidence intervals for a publication 25 years ago, did the calculations myself and got it through peer review.