Wednesday, September 15, 2010

Does "seasonal adjustment" have any place in real estate stats?

I received an e-mail from a concerned reader regarding the Canadian Real Estate Association's most recent claim that sales volumes are up across Canada, led by BC and Ontario. Alarm bells went off immediately when I read BC as we already know that Victoria's sales volume number for August is a 20-year-low. How could that mean that sales were up from July when sales were actually down almost 20% month-over-month?

Are sales booming in other parts of the province? No.

How could this possibly be reported as a positive statistic?

Enter seasonally adjusted data, defined as:
Observations over time (time series data) modified to eliminate the effect of seasonal variations.
Time series data is exactly what we get each month from the VREB: monthly sales statistics. Seasonal variations in the case of local real estate pertain to changes in inventory levels (total active listings, sales volumes and new listings volumes) caused by the seasonality of the business.

Is real estate a seasonal business? We know that Victoria has a pronounced spring buying and selling season as represented by decades of sales statistics. You could make the argument that there is seasonality to the business, much like tourism, because there are months that are traditionally more busy than others.

We also know that when you compare data from say, August 2009 (the month) and August 2010, you're dealing with data from the same month and effectively the same season. Do you need to seasonally adjust the data? No. How about July 2009 with July 2010? No. July 2010 with August 2010? If you believe the summer begins or ends in August, maybe. But really, no.

So why would the CREA and BCREA report sales statistics this way? Because it puts a positive spin on a very poorly performing real estate market.

Different months out-perform other months. You can compare one month to the next, one month of a given year to the same month in previous years and sub-sections of data (e.g. April, May and June) from one year to another. I'm not a statistician or a math whiz by any stretch of the imagination but from a cursory glance I just can't see why you'd want to or need to seasonally adjust real estate sales volumes/dollar-amounts unless you are purposely trying to employ a statistical trick to make data appear favourable.

I know there are several regular commentators here that know far more about stats than I, so I'll ask them to weigh in on this issue in the comments.

27 comments:

DavidL said...

50% of statistics are made up (including this statistic). ;-)

Comparing real estate stats for a specific month over a range of years appears to be an accurate way to show marketplace trends. Applying historical data to attempt to "normalize" a trend (i.e. apply a seasonal adjustment) that is anything-but-normal is a deceptive practice. The past does not predict the future ...

a simple man said...

The simplest answer is that they analysed the dataset until they found what they wanted. It is a valid analysis that is pretty much easiest done by summing up all the sales in the year and dividing by 12 to get a monthly deseasonalized average. There are fancier regression methods to practically get to the same place, but the differences will be minimal.

Obviously comparing monthly data right now to past years' identical month makes the situation seem dire as does comparing to the previous months as of late. The RE industry obviously has changed the way they report data from previous trends to hide the truth - the RE market is correcting and correcting quickly.

They are also not mentioning the total listing numbers or the MOI.

Honesty is the best policy.

Bubble 'n Fizz(le) said...

The simplest answer is that they analysed the dataset until they found what they wanted.

Yes, it's a common strategy. For example, every September this blog compares July and August numbers and declares the market is tanking. Don't believe me? Look for yourself--Sep-2009 and Sep-2008. Howver, since August 2008 average prices in Victoria have increased 2.9%--a moderate increase to be sure, but certainly matching overall inflation. There is also the uncomfortable fact (for you, that is) that in August 2008 there were 3752 active listings in Victoria--more than the 3474 listings in August 2010 (these are BCREA numbers--not seasonally adjusted).

Obviously comparing monthly data right now to past years' identical month makes the situation seem dire as does comparing to the previous months as of late.

Well yes, because one data point does not a trend make. If you look at August 2010 YTD versus August 2009 YTD, which involves a lot more data over a longer period of time, August 2010 looks like a 13% drop in sales volume but 8.4% increase in average price. Again, BCREA numbers, not adjusted to anything. I'd check back in August 2011 before I made any bold predictions about the market tanking. Right now, HHV is batting zilch.

Leo S said...

@Bubble

For example, every September this blog compares July and August numbers and declares the market is tanking. Don't believe me? Look for yourself--Sep-2009 and Sep-2008.

Apparently you didn't actually read those links yourself. In the 2009 post there is no mention of anything tanking. In 2008, the market was tanking, but the post was very cautious and just presented data without making assertions about it.

There is also the uncomfortable fact (for you, that is) that in August 2008 there were 3752 active listings in Victoria--more than the 3474 listings in August 2010 (these are BCREA numbers--not seasonally adjusted).

Where the heck did you get that crazy number for August 2010?
From the VREB: "The number of properties available for sale at the end of last month declined to 4,356"

August 2010 looks like a 13% drop in sales volume but 8.4% increase in average price

Well obviously. No one is disputing there was a big runup in 2009. The topic is the current conditions, not those of 8 months ago.

Unknown said...

Seasonally adjusted data do have a place in an analysis. But they do not tell the whole story. People (including this blog) always want to see a clear answer to all problems but in the real world you often don't get that.

Often, statistical analysis is terrible because it
(a) gets cherry picked and spun to the advantage of a certain point someone wants to make
(b) is often poorly understood by general public

The bottom line for me is: There is no clear answer as to where this market is going. Prediction of complex systems such as the housing market is very difficult and highly uncertain. So you will be able to find some statistical measure in support of almost every prediction. Interpreting the data in a holistic way is difficult and one reason why we see a wide range of predictions.

Time will tell I guess. As far as I am concerned, I think that the market has some real issues (price to income and so on). How that will translate into a long term trend is impossible to say. One cannot predict future direction of a complicated system. Such is life.

Unknown said...

YTD stats are quite meaningless.

Since there have been such few sales the last few months, the low prices aren't reflected very well in a yearly average.

A better way would be to average out each month's average instead of average all the sales in the year.

Marko said...

Anyone see A-Channel news tonight?

;)

DavidL said...

@Marko

No, I didn't ... what was A-Channel reporting?

a simple man said...

Hi BnF:

Glad to have you back contributing on the blog - had some RE showings recently that kept you busy?

I have only been a part of this blog since Feb, I believe, so I cannot be held accountable to anything that was said before that or to things not stated by myself, so park that argument for a circumstance that it is valid.

1.) Yes, I claimed that the market is falling quickly this fall. Absolutely.

As irrefutable evidence please see Double Agent's excellent graphs on the main blog page - I am sure you can see the near linear decreasing sales volume over the past months or that the combined volume for the summer (June through August) is the lowest in at least the past ten years.

I don't think you can refute numbers provided by your own RE Board.

2. I haven't performed the calculations, but I can concede to you that since August 2008 average prices in Victoria have increased by about 2.9%.

I don't think any of us here would claim that RE was cheaper in Aug 2010 that Aug 2008. I think a lot of people would agree that there should have been a decrease in house values after the 2008 recession but due to lax mortgage regulations and historically low interest rates we did not experience much of a drop in prices. Since the rules have since tightened and interest rates are rising we are now seeing what started in 2008 but was interrupted by emergency govt intervention.

3.) There is also the uncomfortable fact (for you, that is) that in August 2008 there were 3752 active listings in Victoria--more than the 3474 listings in August 2010 (these are BCREA numbers--not seasonally adjusted).

Sorry, BnF - but you should check your Board's numbers again - you are about 800 low. Moreover, if you again look at Double Agent's graphs on the main page, the Months of Inventory for Victoria is now approaching 11 months - that is an uncomfortable fact for agents such as yourself.

4.) You are right that one data point does not make a trend. Look at the data points since the start of summer and for the weeks so far this month - pretty obvious trend of declining sales - also declining average price, but one has to be careful when using a mean value with such a small sample size as outliers tend to have a large effect on the reporting of averages (BTW, that will be the spin soon when the low, low sample size of Sept or Oct or Nov is skewed in a kurtotic fashion with the absence of low-end purchases and a few big mansion deals).

I think you can agree that the trend since the early summer has been decreasing sales and average price. And the worst sales volume in 20 years last month, and likely this month again.

5.) I'd check back in August 2011 before I made any bold predictions about the market tanking. Right now, HHV is batting zilch.

Gladly. All I know is that from our last bet made last spring, I don't have to buy you that beer I wagered or eat the humble pie I promised to.

And you?

Have a great night.

a simple man said...
This comment has been removed by the author.
a simple man said...
This comment has been removed by the author.
a simple man said...

and BnF, I am sure you have noticed price reductions of up to $150,000 on houses below $1 million for SFH and many, many dramatic reductions on new condo developments (up to 40% off). In fact, not having a price reduction is becoming a bit of a rarity at the same time that sales under appraisal are becoming common.

When was the last time that occurred?

Please, let's just stick to the facts and not vague statements.

Chris A said...

HHV, deseasonalizing time series data is extremely common and uncontroversial. For example, if you use Statistics Canada's Cansim data you'll have to go out of your way to find non-deseasonalized versions of many time series, because for most purposes the deseasonalized versions are of more use. You can find discussions of seasonal adjustment in any advanced time series econometrics textbook.

As I noted elsewhere, seasonal adjustment amounts to a formalized, more nuanced version of what you say you do informally: compare, say, August 2010 to August 2009, and previous Augusts. We think the market is down now not because sales were lower in August than in April, but because the decline in the interim was steeper this year than in previous years. If we want to know whether sales are higher than usual this August, we mean higher than lower than a usual *August*. That's all deseasonalizing does.

This August was, in fact, worse than previous Augusts in many regions. I already presented analysis showing quite formally that in several regions August 2010 had statistically significantly lower sales than we would expect *after* I corrected for seasonality. And after I also corrected for trends, that is, took into account that sales might just slowly go up or down over time through good and bad markets. I would suggest you insist that the methodology used to conclude otherwise is presented in detail.

ArtVandelay said...

In case you missed the A Channel News tonight.

Check out their YouTube Channel.

DavidL said...

@Chris wrote: deseasonalizing time series data is extremely common and uncontroversial. For example, if you use Statistics Canada's Cansim data you'll have to go out of your way to find non-deseasonalized versions of many time series, because for most purposes the deseasonalized versions are of more use.

Using a seasonal adjustment for time series data that has a cyclical nature is common practice and makes sense. For example, unemployment statistics are seasonally adjusted as employment rates follow a well-defined seasonal fluctuation.

Trying to use seasonal adjustment to analyze trends in markets (whether stock or real estate) hasn't proven to be too effective over the past 80+ years. Just a few years ago, most bankers and analysts were claiming that nobody could predict the dramatic downturn in world markets. Perhaps they were too busy trying to rationalize the numbers and make them fit existing trends rather than more seeing them for what they actually were.

Seasonal adjustment of current real estate trends suggest that the market is stable. The market is dictated by how much sellers are willing to pay. If sellers are currently "holding off" because they think that prices are falling - sellers who need to sell will have no choice but to reduce prices. It becomes a self-fulfilling prophesy. No amount of seasonal adjustment will change this.

ArtVandelay said...

Marko you made the news! Nice work.

DavidL said...

@hammertown: Thanks for the link.

Marcos - I hope you get some business from your "15 seconds of fame". All in all - the report was a balanced piece of journalism.

Marko said...

Total surprise to be on the news. Received the phone call this morning. I definitely need to buy some presentable clothing for future occasions.

The 70% cash back gig is finally gaining some momentum, I have done a few deals and I have a few more people I am working with. I will probably need to wait until spring to see if my business model is going to succeed or not. I need a large volume of deals to make it work for me, probably 3 to 5 per month depending on sale price.

Still struggling to find leads for listings!

jesse said...

Seasonal adjustment is effectively eliminating a known modulation on the longer term trend. There is nothing wrong with it to show long term movements in price.

That said, there is good evidence people do not use seasonal adjustments when deciding what to pay for housing. That prices are generally weaker in the second half of the year and into the first part of the new year should be actively tracked, not canceled, as it shows distortion. That is, it's a bit odd that a multi-decade capital purchase would be so dependent on predictable month-to-month swings in price.

There may be a time where there will be little to no seasonality in house prices.

HouseHuntVictoria said...

Marko, while you were plugging yourself did you plug this blog? (I have a sneaky suspicion here's where they found you, though I could be, and very likely am, very wrong).

BnF: "Right now, HHV is batting zilch."

If I made predictions this statement might be true. But since I rarely make predictions, I'll give you the coles notes of what I do say, ad nauseum:

1. Victoria real estate is over priced when compared to Victoria incomes and Victoria rents

2. Victoria real estate is over priced when compared to Victoria incomes and Victoria rents

3. Whatever the sales volumes, active listings and average prices are reported by VREB each month

4. Victoria real estate is over priced when compared to Victoria incomes and Victoria rents

5. The MSM republishes the VREB/CREA/BCREA press releases without writing much in the way of balance and certainly never doing any "investigative journalism"

6. Victoria real estate is over priced when compared to Victoria incomes and Victoria rents

7. You read what you want to read, not what I've written

a simple man said...

Wow...balanced news reporting? I have a new favorite local channel.

Good for you, Marko. And good for A Channel for using real numbers.

a simple man said...

Hey Marko - are you already working with a new RE company?

What happened with Sutton?

Marko said...

Simple man,

The new company I am with has a much lower overhead, like 60-70% lower. This allows me to pursue my cash back business model with a greater chance of success.

Certain agents drive high-end cars, I drive a Honda Civic. At some real estate companies you have access to really nice offices, at others you don't. Lower overhead means I can pass more onto my clients.

Marko

Chris A said...

DavidL, forecasting is a different exercise than seasonal adjustment.

Stock prices and many other similar time series follow stochastic processes which have the property that changes in these series tomorrow are impossible or extremely difficult to forecast today. That means, among other things, that they do not have seasonal components, nor do they have trends.

Real estate unit sales are not like stock prices: they have predictable seasonal components, much like employment statistics. And just employment statistics, if we want to know whether the market is currently bad or good, we need to adjust for seasonality. That's the reason Statistics Canada commonly presents such statistics after deseasonalizing. That's also the reason I deseasonalized data on Victoria real estate sales. It isn't voodoo, it isn't forecasting, it's a bog-standard and entirely uncontroversial procedure. If you want to learn more about it, again, the relevant textbooks all contain lengthy discussions.

I think you misunderstand what deseasonalizing is, fundamentally: it *is* a formal way of "comparing real estate stats for a specific month over a range of years." If you agree that we should compare like to like, in the sense that we should compare this August to previous Augusts instead of to, say, this April to figure out whether the market is up or down, you agree we need to do something to deseasonalize. There are better ways to do that than just eyeballing a graph, or crude and inefficient methods like a twelve month avergage. Deseasonalizing is not forecasting, and it is not "normalizing" data, which is something else entirely (and also an entirely uncontroversial and common procedure).

DavidL said...

@Chris

You have written your point of view well - but I don't agree.

The real estate market is much like the weather. They both share:
* volumes of historical data
* seasonal variances
* limited ability to forecast

Using the weather example ... predicting trends in weather events one month into the future is barely possible. Predicting specific events within a few days is possible. Using a "seasonal adjustment" to analyze weather data provides little insight as variances in trends are obscured by the adjustment process.

Back to real estate ... what is the purpose of using a "seasonal adjustment" other than to try to explain the current/historical market conditions and predict future market trends. Adjusting the data implicitly suggests that analysis will be performed in order to determine such conclusions. Using a "seasonal adjustment" to analyze real estate data provides little insight as variances in trends are obscured by the adjustment process.

Chris A said...

DavidL, I just got through explaining, again, that deseasonalizing a time series is not the same as forecasting. I don't know how to make the point in yet another way: deasonalizing is not forecasting; responding to the claim that deseasonalizing is a standard and useful procedure by pointing out that it's hard to forecast, as you have now done three times, is a non sequitur. As I have repeatedly explained.

On weather: If we wanted to know whether this past Summer was a good or a bad Summer, we would compare this Summer to previous Summers, not this Summer to this Winter. The weather has a predictable seasonal component, and clearly we should take that into account when we evaluate whether this Summer is relatively bad. When researchers try to estimate the extent of global warming, for example, they take into account seasonality in weather.

Similarly, we should take the seasonal component in sales into account when we try to assess the current state of the market. Yet again, evaluating what has happened does not imply we're trying to forecast what will happen: I explained at some length on VV why forecasting real estate prices is at best extremely difficult, and why we ought not place much confidence in extrapolating local trends.

And at the same time, I maintain that deseasonalizing time series like house sales in Victoria is a common, uncontroversial, and useful procedure. Again, you can learn about these issues in any time series econometrics textbook: this is standard, elementary stuff.

Alexandrahere said...

You can analyze, theorize and intellectualize all you want about what is, was and will be in terms of the real estate market here in Canada, BC and in particular Victoria.

However the fact is, given the average household income (or was that median household income reported in the times/colonist recently?), in Victoria is almost $78,000 (the 8th highest in Canada); that household can only "afford" between 3 - 4/1 yrs income to pay for the family home. If the average family home here costs $600,000 then people are paying 7.69 years of their income for the price of their home. Simply put, this is not sustainable. We all know why it has been i.e. due to government interference and blah, blah, blah, however since around 2005 it has not been a question of IF the market will bust, it has been the question WHEN will it bust.

The answer is: The fall of the current real estate market started in 2008, got lifted back up again in 2009 by we know who and why, and it started spiraling down again in the spring of 2010.

When people believe something to be true....it is true. People believed, as they have in the past that real estate prices can only go up and they rushed in to get their slice of pie. And they ate and licked their lips and even added some ice cream to it. But things change. People now believe (for a number of reasons), that real estate is going to tank....and so it will as the herd of cattle have made an about turn and it is trampling all over the market place, leaving a path of devastation and ruins for the many left behind.