Jump to content
House Price Crash Forum
Sign in to follow this  
Levy process

What Is The Point Of Seasonally Adjusted Price Changes?

Recommended Posts

What is the point of seasonally adjusted values, such as monthly house price changes, if even the seasonally adjusted value fluctuates about a lot (like between -0.3% and +1.4% for example) within the space of a year?

Surely the whole point of a seasonally adjusted figure is to smooth the data out so that the figure quoted on a given month is representative of the total change expected over the year, isn't it? Surely the point is to shift the monthly figure up or down in proportion to what the expected pattern is every year, such that at the end of a complete year, the monthly figures are all roughly the same, or at the very least, change smoothly over the year as part of a long term trend?

Having a huge scatter of seasonally adjusted figures seems to indicate to me that the seasonal adjustment formula is just wrong, because it is not producing the effect that it is intended to produce, namely to keep all the figures close to the longer term trend.

Or is it just me?

Share this post


Link to post
Share on other sites

... for example, try doing a calculation of the RAW (not adjusted) monthly changes in house price using the Nationwide's data on average price. I came out with a range of values which ranged from -0.65% to 1.46%. That's hardly larger than the spread on their quoted seasonally adjusted figures. So what does the Seasonal adjustment add to the picture?

Share this post


Link to post
Share on other sites

... for example, try doing a calculation of the RAW (not adjusted) monthly changes in house price using the Nationwide's data on average price. I came out with a range of values which ranged from -0.65% to 1.46%. That's hardly larger than the spread on their quoted seasonally adjusted figures. So what does the Seasonal adjustment add to the picture?

Seasonal adjustment is not intended to "smooth the data out so that the figure quoted on a given month is representative of the total change expected over the year", but to take account of the fluctuations that would normally be caused by the variation in activity at different times of the year in order to get at the "true picture" of what was happening in that particular month. So there's no reason at all to expect the monthly adjusted figures to have less of a spread than the unadjusted. Though it is true that the seasonal adjustment creates some slightly bizarre effects as different years have very different patterns.

In fact, a month is probably just too short a period to be statistically relevant to HPI - the YOY figures are more use, but we're all so obsessed with house prices, we end up poring over the MOM and moaning about seasonal adjustment etc...

Share this post


Link to post
Share on other sites

Seasonal adjustment is not intended to "smooth the data out so that the figure quoted on a given month is representative of the total change expected over the year", but to take account of the fluctuations that would normally be caused by the variation in activity at different times of the year in order to get at the "true picture" of what was happening in that particular month. So there's no reason at all to expect the monthly adjusted figures to have less of a spread than the unadjusted. Though it is true that the seasonal adjustment creates some slightly bizarre effects as different years have very different patterns.

Not sure I agree with that. Of course the point of it is to take account of the normal seasonal patterns. But if there is any justification for a "normal seasonal pattern" then it must surely be because over recent years, the pattern has been the roughly the same. If it hasn't been roughly the same, then there IS NO SEASONAL PATTERN, and so you can't do a meaningful seasonal adjustment.

I think that in general, for an index that has a long term trend with a predictable seasonal short term pattern superimposed on it, then by doing a seasonal adjustment, you will end up with a value that represents the smooth averaged total-year change. It think if you don't, then it means the situation is one where you cannot apply a seasonal adjustment meaningfully.

Consider this thought experiment:

Imagine prices vary showing the same seasonal fluctuation every year, superimposed on a steady annual growth rate of 6%(the actual value doesn't matter). So after a few years, a seasonal adjustment factor is applied to the monthly reports, to take away the repeated seasonal changes. Now imagine you do that year after year, and every year the annual change is 6% (totally hypothetical example here). So what you would require after the seasonal adjustment is a mothly value of 6%/12 = 0.5% surely?

If in my example above after the seasonal adjustment you DON'T get a constant value every month, but instead still get fluctuations on a monthly basis, but you get year after year a total year change of 6%, then you're GETTING THE SEASONAL ADJUSTMENT WRONG, surely?

Of course in real life the annual change varies over the years, but you can see my point? If the true picture of a market is of changes in annnual growth rate on long timescales (i.e. years) then ANY large monthly fluctuations must be "seasonal" effects, OR there is no such thing as a seasonal effect that can be removed.

What I am trying to say is, EITHER the monthly fluctuations are repeated over several years, in which case they are TRUE seasonal factors that can be adjusted for to leave a value representitive of the longer term slower changes in rate, OR you are in case where the actual market is really fluctuating on a monthly basis in new ways which it doesn't repeat every year, in which case the concept of "seasonal adjustment" makes no sense, because you are factoring out what happens every year, but at the same time saying it isn't happening this year.

At the very least, if you apply seasonal adjustment and get values that are not representitive of the longer term trend, then what you have discovered surely is that there is currently NO long term trend, because the market has departed from that, and you are in a period of wild fluctuations which are NOT the usual seasonal effects (i.e. they are fluctuations not normally seen).

Or have I missed the point...?

Share this post


Link to post
Share on other sites

Is nobody else interested in this question? I think it is quite relevant to the house price data that people on here are so fond of criticising. In a nutshell - what is the justification for the "seasonally" adjusted data that is presented, given that it produces no indication of anything smoother and closer to a long term trend than the raw data, and therefore indicates that there can be no seasonal effects repeating this year that have been occurring last year?

Share this post


Link to post
Share on other sites

I think that in general, for an index that has a long term trend with a predictable seasonal short term pattern superimposed on it, then by doing a seasonal adjustment, you will end up with a value that represents the smooth averaged total-year change.

No - imagine a year in which prices go up 2% a month for six months, then down for six months back to the starting point. The monthly rate, seasonally adjusted or not, should show something like 2% rises and falls, even though the annual rate averaged over the year is 0%. If monthly rates are statistically relevant (which I doubt), it's to give you a clue to the direction and rate of change in the annual rate.

Consider this thought experiment:

Imagine prices vary showing the same seasonal fluctuation every year, superimposed on a steady annual growth rate of 6%(the actual value doesn't matter). So after a few years, a seasonal adjustment factor is applied to the monthly reports, to take away the repeated seasonal changes. Now imagine you do that year after year, and every year the annual change is 6% (totally hypothetical example here). So what you would require after the seasonal adjustment is a mothly value of 6%/12 = 0.5% surely?

No this thought experiment only gives that result because you are assuming a constant rate of change. In any other situation it's not that simple

At the very least, if you apply seasonal adjustment and get values that are not representitive of the longer term trend, then what you have discovered surely is that there is currently NO long term trend

I don't agree. I agree (as would a proper statistician I'm sure) that seasonal adjustment can be a bit of a clumsy tool, but there are genuine regular fluctuations through the year and applying seasonal adjustment consistently is probably as good a way of trying to account for this as any.

My main point is not about the detail of this anyway. I think we all get too obsessed with daily and monthly changes, and when they don't seem to go our way we try to find detailed fault in the figures. YOY figures are now varying between slightly +ve and slightly -ve on different indices. The important thing is where they go next, not why the monthly figures might be fractionally different if we used a different statistical method.

It's the wood, not the trees...

Share this post


Link to post
Share on other sites

The idea behind seasonal adjustment is to decompose a fluctuating series X(t) into separate components T(t)*S(t)*I(t), where T(t) = trend, S(t) = seasonal variation, I(t) = irregular part. The aim is to extract S and then divide it out from the series leaving just the T*I bits, i.e. all those effects *except* those down purely to the time of year; and knowing the trend T and the irregular variation I *uncorrelated* with time of year is also quite handy.

There’s an Exel file that demonstrates the standard X11 method applied to the decomposition of some unadjusted Land Registry data; see this thread:

http://www.housepricecrash.co.uk/forum/ind...ndpost&p=225761

Share this post


Link to post
Share on other sites
Guest wrongmove

In a nutshell - what is the justification for the "seasonally" adjusted data that is presented, given that it produces no indication of anything smoother and closer to a long term trend than the raw data, and therefore indicates that there can be no seasonal effects repeating this year that have been occurring last year?

If you go to the graphs section of HPC.co.uk, the top graph is transactions. Each year, the transactions flucyuate in a pretty regular way. Without seasonal adjustment, you would have to know the typical stats for every month individually, to account for the periodic swings in transactions.

If the figures weren't SA, we would just have to do it ourselves. Otherwise, a bear, for example, would just be happy every autumn (when transactions are low) and fed up every spring. The SA allows us to see past these yearly swings to see if anything interesting is really happening.

But of course SA is far from perfect, but look at the graph and its obvious why it is done.

http://www.housepricecrash.co.uk/house-pri...rash-graphs.php

Share this post


Link to post
Share on other sites

The idea behind seasonal adjustment is to decompose a fluctuating series X(t) into separate components T(t)*S(t)*I(t), where T(t) = trend, S(t) = seasonal variation, I(t) = irregular part. The aim is to extract S and then divide it out from the series leaving just the T*I bits, i.e. all those effects *except* those down purely to the time of year; and knowing the trend T and the irregular variation I *uncorrelated* with time of year is also quite handy.

There’s an Exel file that demonstrates the standard X11 method applied to the decomposition of some unadjusted Land Registry data; see this thread:

http://www.housepricecrash.co.uk/forum/ind...ndpost&p=225761

Thanks, that is a very clear description.

A further question: If the I part are fluctuations of roughly the same magnitude as the S part (but of course occuring at different times), then how many years would have to pass before the S part could no longer be said to be present?

Share this post


Link to post
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

Loading...
Sign in to follow this  

  • Recently Browsing   0 members

    No registered users viewing this page.

  • 301 Brexit, House prices and Summer 2020

    1. 1. Including the effects Brexit, where do you think average UK house prices will be relative to now in June 2020?


      • down 5% +
      • down 2.5%
      • Even
      • up 2.5%
      • up 5%



×
×
  • Create New...

Important Information

We have placed cookies on your device to help make this website better. You can adjust your cookie settings, otherwise we'll assume you're okay to continue.