Tribeca Co-op vs Condo Market / Flaws of Small Sample Size
A: Wanted to have a quick discussion about a common problem that I often encounter when discussing UD real-time data tools with fellow brokers and clients. Lets use the Tribeca market as an example so the discussion can be a market update as well. Every broker out there would like to be able to provide accurate information on exactly what their client is buying or selling into. For example, what is happening in the Tribeca, $700,000 to $850,000 price range, 1br/1bth, postwar full service doorman condo market that allows pets and has a building roofdeck. So they ask me if I can add in that functionality but I try to explain that if they get too granular in their request for real time charts on a Manhattan submarket, that there will not be enough data to push forth a worthwhile trend - the result could be very high % change numbers and a useless chart that either under or over-exaggerates what is really happening out there. For me the debate has become whether to build an interface that only goes so far but still pushes forward the highest quality trend OR let the user chart whatever they want and know that 1,000s of useless chart combinations exist. Lets discuss.
Being in a service industry, Manhattan brokers want to give their clients what they ask for; and who can blame them. From building policies and amenities to manually entering in a small price range. But when it comes to charts on this marketplace, it's best to focus on quality than granularity. For those that don't know:
Granularity is the extent to which a system is broken down into small parts
For Manhattan charts this would mean the extent to which we allow you to slice & dice the Manhattan market to see whats happening (think bedrooms, bathrooms, postwar/prewar, apt size range, price range, bldg type, bldg service level, roofdeck, building policies, etc.)
Sample size refers to the number of objects in the sample
For Manhattan charts this simply means how many units are either active, pending or off-mkt.
For our interface on UrbanDigs you can't let the user get too granular or else you will risk having too many charts with very low sample size rendering the trend useless.
Lets check in on the Tribeca marketplace and compare pending sales trends for both co-ops and condos:
Some facts:
Fact #1 - This is showing you Demand trends (Pending Sales) for the entire Tribeca Co-op & Condo Market!
Fact #2 - There are only 5 Tribeca Co-ops Pending right now!
Fact #3 - Over the last 2 years, Pending Sales for Tribeca Condos are +8.7%
Fact #4 - Over the last 2 years, Pending Sales for Tribeca Co-ops are -45.5%
Now lets see what happens when you want to see only the Tribeca Co-op market with 2+ bathrooms and priced between $1M-$2M (getting more granular):
Only 1, and the chart is basically useless. The trend went from 4 to 1 and is down 75% over this time period. If 1 more unit goes to contract, it will go to 2 and show a quarterly rise of 100%. Useless.
Interpretations might be:
Interpretation #1 - Tribeca 2+ Bath Co-op Market must be really weak
Interpretation #2 - Tribeca 2+ Co-op Market likely has little to no supply
I understand this is a section of Manhattan that we all know has hardly any inventory, but that is the point of the discussion.
This is the reason why we have chosen not to add the following details to our chart interface (at least at this time):
- building service level (f/s drmn, p/t drmn, elev only, walkup)
- pet policy
- private outdoor space
- prewar/postwar
- smaller price ranges
- building amenities (roofdecks, pools, storage, etc.)
- monthly carry maximums (maintenance + re taxes)
- flip tax (yes, I even get asked to be able to chart out only buildings with flip tax)
etc..Imagine how many chart combinations will be useless..you will try to give the client exactly what they want but you'll have no idea what it really means in the end? The point is to build a system that gives you customizability and does not sacrifice showing you meaningful trends.
For this case, I would advise my client to focus on the broader Tribeca trend when discussing buy side or sell side strategies since we have significantly more data on it.
Would love some thoughts on the topic...
Posted by krishurd17
Fri Jan 13th, 2012 08:44 AM
Noah..I could not agree with you more. One transaction does not make a trend. It could reveal more about demand in a area, but not a real trend. Ie 1 pending in tribecca vs 10 pending in another area, using same criteria. If you were to parse by anything granular, I would consider only attended vs non-attended. Just a thought. :)
Posted by urbandigs
Fri Jan 13th, 2012 09:13 AM
hmmm, can you explain that last part please Kristin? many thx!
Posted by krishurd17
Fri Jan 13th, 2012 04:39 PM
I have found that my clients will pay more for a doorman building. As you so eloquently explained above how too much criteria would ruin then data, but a factor such as this one may actually effect price trends a bit. For example, someone on lower fiftth would be willing to pay more to be in a doorman building rather than non-doorman buildings.
As a result, the two markets a different, very different. Make sense? This may be limited to pricing trends only, but a factor.
Thoughts?
Posted by urbandigs
Fri Jan 13th, 2012 05:05 PM
Aha, I was overthinking it..thought you were refferring to a methodology change outside of adding another input. Gotcha! Yes I totally agree. I think we can work this into the comps system as an option when comparing nhood price trends..thx as always!
Posted by krishurd17
Fri Jan 13th, 2012 09:27 PM
sometimes, it's just a simple answer:)
Posted by Timo
Sat Jan 14th, 2012 11:32 PM
Why not just add error bars? The fewer the data points, the bigger the error bars, to the point that it may be much bigger than the actual data.
Posted by urbandigs
Sun Jan 15th, 2012 08:50 AM
Timo - yes we considered that, but its just one more part of the chart that the user now has to understand. All anyone really cares about is the 'interpretation" of the charts trend - to me, as I do training sessions at offices for different manhattan brokerage firms, this is by far the biggest challenge to us going forward. Creating an easy way to tell us what the charts are saying. If we got more granular in the interface, it would degrade the whole idea of only pushing forward worthwhile charts to users.
Back in 2010, we were thinking a static 'confidence' barometer on every chart page that signals the quality of the trend its showing you, and telling you what the logic is as to why its high or low (i.e., not enough data, highly volatile swings in past 2 quarters, data is exposed to 3-6 months lag, etc).
But in the end we decided not to build/implement the idea. We r working on another approach that I think will make this concern go away...i hope! thx for commenting!