On Thursday, the New York Times published a page one piece by the Upshot’s Josh Barro making the claim that faced with competition from e-hailing car service apps Uber and Lyft, taxi medallion prices are “plummeting.” The basis for the claim, from what I can tell, is largely this (in the second graf):
The average price of an individual New York City taxi medallion fell to $872,000 in October, down 17 percent from a peak reached in the spring of 2013, according to an analysis of sales data.
Sounds like an alarming decline. But the article goes on to say that these averages are based on very few observations per month. Taxi medallions are not bought and sold very often; the Times notes there was just one medallion sold in September (but nine in October). Here’s the Times graphic:
To be sure, the article offers a few other data points mostly in the form of anecdotal evidence and isolated instances of difficulties medallion owners and taxi commissions have faced selling and auctioning medallions in recent months. But I was curious whether the decline the Times was hyping was even statistically significant given such a tiny number of observations. It could very well be caused by regression to the mean. After all, just a few years ago, medallions were selling for a fraction of the price. Using annual data, the Taxi and Limousine Commission (TLC) shows medallion prices have increased at a rate that looks exponential in recent years:
The good news is that the TLC data is publicly available going back to 2009. The bad news is that it’s in an extremely annoying format, with each month’s transfers listed in individual PDFs. Here’s what the October data looks like:
Notice a lot of complicating factors. Some sales are in stock, rather than assets. Some are corporate sales rather than individual sales. Sometimes reasons are listed. Sometimes not. There are a number of transfers for which no sales price is listed (the article indicates these and other below-market transfers are often between family members).
Still, I forged ahead and collected the monthly sales data going back to January 2012 (would love to do the rest but I simply didn’t have the patience). Looking just at transfers excluding fleet owners (as The Times did), and excluding all transfers at $0, there were just 287 transfers between January 2012 and October 2014 of either medallion assets or stock. That’s not a lot of data to make any bold claims about prices “plummeting.” But this is what the time series looks like, using sales of partial shares of medallions as observations implying a valuation of medallion prices. The number of observations in each month are plotted at the bottom.
While some decline is certainly evident in recent months, is this decline out of the ordinary, solid evidence of an actual new trend? Far from it. Indeed, if anything the scatterplot indicates some potential seasonality in the data, with declines in precisely this time of year annually. (For some reason, October 2012 data was missing online, making a more thorough time series analysis of seasonal lags difficult). In terms of the number of transactions per month, perhaps a more telling bit of evidence, there does appear to be some modest decline but again hardly an unprecedented drop-off.
Meanwhile, if you look at all medallion transfers (rather than excluding those transactions with fleets), the decline is even less evident. And considering that a non-trivial number of transfers by individuals listed in the data are transfers to corporate buyers, it’s not obvious why we should look only at one or the other.
Again, we see some decline but hardly the “tumbling” prices Barro describes. Is the data flawed? Sure, for all the reasons recounted in the article, including the important but difficult to detect role played by the institutions that provide financing for medallion transfers. The bottom line remains that it is really hard to assess the going rate for medallions when the market is so thin and noisy. To slap a headline on these far-from-conclusive data announcing on the front page that prices are “plummeting” just seems irresponsible journalistically.
Why does any of this matter? Well, word is that Uber is close to securing yet another round of funding and a $40 billion valuation. These sky-high valuations are based on the notion that it will someday conquer a vast share of the for-hire car service and traditional taxi industry, and the NYC market remains a significant piece of that pie. For Uber-boosters, it’s an article of faith that these “disrupting” apps are fomenting a “revolution” in the taxi business (as Barro puts it); they are tapping into and creating vast new markets for rides. (Critics like NYU Finance Professor Aswath Damodaran have argued that these assumptions may be flawed and cautions that plans to expand into other kinds of delivery services are hardly a sure bet.)
My point is simply this: the highly regulated and competitive NYC taxi industry remains difficult to crack. While in some cities, like Chicago, one of the anecdotal cases Barro focuses on, UberX really is cheaper than the existing medallion taxis, but that may not actually be true in New York (despite Uber’s well-publicized claims). UberX and Lyft (and whoever else) present a real competitive pressure on the traditional taxi industry, but it’s not as though taxi drivers who own or lease medallions are suddenly in possession of worthless assets. In Manhattan, where most taxi rides originate, taxis remains plentiful and they aren’t going away. The ease of hailing a ride by tapping on a phone versus hailing a cab on the street, the old fashioned way, hardly justifies the breathless coverage surrounding their supposed effects on the future of transportation.
Medallion prices may well be in decline, and given the spike in prices over the last decades, that’s probably a good thing. But the empirical evidence doesn’t quite match the theory, and it would serve the Times to be a little more skeptical of the claims and agendas of companies before repeating their favored narratives.
UPDATE: Josh Barro was critical of my second scatterplot above because the individual and mini-fleet medallions are technically different classes of assets, with different regulations associated with them (see page 12 here, if you care). The distinction matters, although if the trend is the same and the competition from apps like Uber is impacting both groups, I’m not sure it should matter as far as detecting what’s actually happening in the data. Nonetheless, here’s the two groups broken out separately. The average for corporate medallion values does appear to fluctuate more wildly, but that’s because there are so few transfers (with prices) during several periods in the data. Sometimes the “average” is based on just 1 transfer; in many months there were no transfers with market prices. That only underscores my broader point; looking at average values in these data doesn’t make much sense if you want to demonstrate a real plummet in prices. This transfer data is both noisy and sparse, which is why it isn’t a good idea to make too much of a particular 17 percent recent decline in the average last month. The number sounds precise and authoritative. It’s not.