Is it a good time to buy a house in San Francisco?

Real estate in SF is up 33% since its low in March 2009.  To someone living in the Bay Area thinking about buying a house that seems daunting. And it doesn’t help that the likes of Twitter, Dropbox, Snapchat and Facebook have crazy valuations.

This post is my attempt at getting confidence in these questions:

1)   How much appreciation can I expect if I have to buy now and decide to sell in 5 years?

2)   If we are in a mini tech bubble and the bubble pops, what will be the impact on housing prices?

Model basics

My regression model consisted of the following:

1)   Supply of new homes – Privately Owned Housing Starts Authorized by Building Permits: 1-Unit Structures for San Francisco-Oakland-Fremont, CA (MSA) SANF806BP1FHSA

2)   Nasdaq as an indicator of how technology was doing (my assumption here is that a lot of the growth and decline in employment/salaries in the valley in the bay area is driven by tech); appreciation in the Nasdaq also incorporates asset inflation

3)   Long term interest rates – 10 year US treasury note to reflect what’s happening in mortgage rates; one flaw is that I haven’t taken into account the risk premium associated with mortgages which is also changes.

These are used to priced the home price index for SF (~25 years of data). Here is a plot of all the variables indexed to 100.

Housing prices, interest rates, nasdaq, and supply growth over last 25 years
Housing prices, interest rates, nasdaq, and supply growth over last 25 years

One thing to note here is that when the tech bubble burst housing prices slid only 10%. So that answers the question about if there is a bubble in social media the impact on housing prices may not be severe necessarily.

Does the model work?

Kind of. The model outputs show that these variables explain about 50% of the variance in the housing market and its significant:

Regression model outputs: Housing prices regressed against supply, interest rates, nasdaq;
Regression model outputs: Housing prices regressed against supply, interest rates, nasdaq;

Where does it work and where does it not?

Times when predicted  actuals buy - fundamentals are on your side
Times when predicted < actual mean don’t buy. When predicted > actuals buy – fundamentals are on your side

Model correctly predicts that home prices should not have risen in 2004-08 era. This means that given interest rates, Nasdaq, and supply the market was valued higher than these fundamental indicators would predict. If you bought then you would be taking a risk that values would adjust back to what was suggested by these indicators and you could lose money.

So now we can develop a measure of how good of a time it is to buy a house: when the prediction is much lower than the Home Price Index, don’t buy prices could fall (the red areas in the graph above). When the prediction is much higher than the Home Prices Index, homes are undervalued. Time to buy (the green areas).

Testing this new iteration

Let’s test how accurate this would have been. Take a new variable which I will call Delta =  Predicted prices / Actual prices. So in times when Delta > 1, you are getting a good deal (green areas in the graph above).

Now we plot Delta versus the 5 year forward return on home prices (unlevered) if you had invested in a home at that time (see Figure below). For example if you bought in 1988-01-01 (the first point), Delta would be 1.15 so it would be good time to buy (Delta >1). So if you bought then the red line represents that if you bought in 1988  your home would have appreciated about 1.3x or 30%.

When Delta >1 historically home prices appreciated over 5 years (and vice versa)
When Delta >1 historically home prices appreciated over 5 years (and vice versa)

The correlation between the measure of Delta and the 5 year appreciation is 84% and further more whenever the 5 year home price appreciation has been negative ‘Delta’’ has been less than one. Thus I can feel good if I buy a home I can predict the 5 year appreciation based on the delta and as long as Delta is above 1 I won’t lose money.

Bottom line

At the end of the graph you will see delta (the blue line) is hovering around 1. Based on where the Nasdaq, 10 year rates, and supply is right now, buying now will not yield to you a big upside in 5 years. You could wait to buy when Delta is >1 but it could take 5-7 years. If you can’t wait that long,  find something you feel is a relatively good deal but be picky.


Update: Excel model incase you want to change / update.

Are Amazon Book Reviews Completely Biased?

Is Amazon missing some helicopters?
Is Amazon missing some helicopters?

My ratings obsession

Anyone who knows me knows that I look at a lot of ratings before I decide to do anything. A good friend of mine once joked I wouldn’t turn left if Yelp didn’t tell me to. Have I taken it too far? – probably – I mean I’m writing a blog post on ratings.

Recently, I was looking for a new book to read when I came across Goodreads. I signed up because it has a recommendation engine that can tell  you what you might like based on books you have read and what your friends are reading.

As I browsed the ratings on Goodreads I wondered if they were any different than Amazon’s. What I saw was surprising – books had a very different ratings on Amazon vs Good Reads, even after having hundreds and thousands of ratings. So which one is better?


The Data: Goodreads versus Amazon – Ratings distribution

Based on an index of 10 books equally weighted (weighted towards business books):
These books received 15K reviews for Amazon and 315K reviews for Good Reads

Goodreads versus Amazon - Ratings distribution

How can thousands of reviews have such a different mean on one site versus the other? And why do Amazon’s ratings have such a  “U” shape compared Goodreads?

Alexa stats
Alexa stats

At first thought you might think – different audiences?

Thanks to Alexa we see it’s not age, education, or family status. The only major difference is that Goodreads skews female more heavily than Amazon. But that doesn’t explain it – why would women on average rate books lower and have a more uniform distribution in ratings. It must have something to do with the ratings system itself. It turns out it probably does (thanks to Sid for helping remind me of the bias in #1):

1) The more work one must do to submit a review, the more extremes you will see in the rating (“U” shape) On Goodreads you simply have to fill out a star rating while on Amazon you have to write a full review. Hence you’ll see many more 1s and 5s on Amazon than what might be true in the real world. So when you look at reviews on Amazon, realize in reality there are a lot of people in the middle who think the books are really 3s and 4s who didn’t review it.

2) There are penalties for dishonesty on Goodreads. Goodreads is a recommendation site. To start getting recommendations you have to rate 20 books you have read. If you aren’t harshly honest about the books you stand the risk of getting bad recommendations.
Amazon, on the other hand, is much more outward facing. In a public review of the book, you may say it was good but it wasn’t for you and settle at 4 stars, but you definitely don’t want to read something like that again so it might get 2 stars on Goodreads.
(Amazon has recommendations too – but these are much more skewed towards browsing history)

All this is to say I think Goodreads reviewers have more skin in the game and are a more representative sample. For now Goodreads = better reads.