In response to the forthcoming FRA loan application by XpressWest (the rebranded Desert Xpress) for its high-speed rail line from the edge of the Los Angeles metro area to Las Vegas, Reason published a report claiming the project would fail. Coauthors Wendell Cox, who cowrote a fraudulent report about Florida HSR, and Adrian Moore, argue that costs will be higher and ridership lower than expected, leading to operating losses and bankruptcy. I still have some doubts about XpressWest’s business plan, but Cox and Moore skirt or ignore the real problems, and instead choose to attack it using numbers that are distorted and at times completely made up.

The smoking gun that something nefarious is going on is the attempt to remodel ridership in terms of competition with cars and planes. In table 2 on PDF-page 20, the report shows door-to-door travel times by the different modes to Las Vegas from various origins in Southern California, including Victorville itself, Riverside (80 km and a mountain pass away), and Los Angeles (130 km away). The assumption, which is for the most part correct, is that passengers drive to the airport or train station and need to factor in congestion, and the explicit assumptions on access time are spelled out in table A-1. The zinger is that while station and airport access times are computed by taking the free-flow Google Maps travel time and adding a congestion cushion, the assumed door-to-door travel times for people driving assume free-flow travel – and even this required me to pick a particular (albeit reasonable) location on the Strip that is closer in than the Google Maps point labeled Las Vegas.

For examples, the travel times by car given from Victorville, Riverside, and Los Angeles are 2:56, 3:47, and 4:20. Those are approximately equal to the free-flow travel times to the Palazzo on the Strip. Needless to say, traffic is not free-flow in Southern California. As of this writing, on Friday at 4:15 pm Pacific Time, Google Maps gives me a travel time of 4:23 from Los Angeles to the Palazzo free-flow but 5:13 in current traffic; figure the extra 50 minutes make it 5:10 over the 4:20 given in the study. The door-to-door travel time for a train from Los Angeles is given as 5:04 to Vegas and 4:04 from Vegas, the difference coming from not needing to budget as much time for the possibility of traffic and arrive extra-early. In other words, including realistic rush-hour conditions, driving is not 14 minutes faster than the train on average in each direction, but 36 minutes slower.

In addition, the report slightly overstates the train’s travel time, as 1:40. The environmental impact statement claims, on PDF-page 39 of FEIS chapter 2, that 150 mph electric trains (the alternative that has since been selected) will take 1:24. While this is an ambitious average speed for this top speed, it is achievable for a nonstop train. Subtract 16 minutes from train time and now driving all the way from Los Angeles is 52 minutes slower than the train. As an additional check on the model, Cox and Moore assume travelers must arrive at the train station 20 minutes before departure, in addition to the congestion cushion. This is not observed in HSR systems in such countries as France and Germany, where open station design means people can arrive a few minutes before departure. Figure 5 minutes and now driving is 1:07 slower than the train.

Let us now step back and examine the general argument of the report. Cox and Moore argue the following: there is a tendency for costs to escalate (as examined by Bent Flyvbjerg) and for ridership to fall short of predictions (they call it the International Average Error Forecast but supply no reference and give no indication of the computation involved, and given the above zinger regarding travel time nobody should trust this). The ridership model has flaws, and a series of sanity checks argue that ridership will fall far short while costs will escalate. It is therefore better, they claim, to expand I-15 instead to deal with rush hour capacity.

At every step of the way, the report makes substantial errors. Cox seems aggressively uninterested in the actual causes of cost escalation and ridership shortfalls, following Flyvbjerg’s note in his original paper that cost escalation can come from many sources but it is fairly certain that there will be some cost escalation in a megaproject.

We can do better, and examine recent HSR projects. In Spain, some meet projections and some do not. For example, the Madrid-Barcelona corridor was 25% below projections in 2010, and appears to have fallen farther behind in 2011 – but in 2008 the line was only 4% behind projections, and with a deep recession and 20% unemployment, Spain can be excused for having less economic activity than projected at the height of its bubble. Likewise, in Taiwan and South Korea the HSR lines have fallen far below projections made in the 1990s, when their economic growth was extremely fast – but even those projections failed a sanity check: Korea thought it would get more HSR riders than the Sanyo Shinkansen, which looks reasonable based on city sizes until one remembers that the Sanyo Shinkansen also connects to Tokyo at one end and the KTX does not; Taiwan had estimated similar ridership, even though its largest city, Taipei, had not many more people than the Sanyo Shinkansen’s distant-second largest city and only one third as many as Sanyo’s largest, Osaka. In contrast, French lines tend to overshoot projections, as can be seen in the above link for Taiwan.

In all cases it can take a few years for ridership to build up: Taiwan took 2 years to achieve profitability after depreciation but before interest (and is now profitable even after interest after a refinancing at a lower interest rate), which Cox and Moore spin as “The project suffered an accumulated loss of two-thirds of its private investment in the first 2.5 years of operation.”

Las Vegas did have a bubble, and is slowing down now, although it is nowhere near the level of depression Spain is in. The report in fact mentions that growth in hotel rooms and travel to Las Vegas has stalled (although part of it is due to the national recession, rather than a Nevada-specific crash). It comes close to mark, but even here it fails to note possible similarities and differences with case studies of shortfalls. However, since the report attacks not just projected 2035 growth but also base-case ridership for 2012, it does not deserve this charity, even as here it skirted a real problem rather than completely missing it.

To criticize the actual model, on PDF-page 34 Cox and Moore attack it for surveying a sample of 400 people and asking them if they would ride the train. They attack the general approach of stated-preference, without giving any reference for why it is bad (they include one sentence of criticism), and then offer the following platitude: “It would seem that a prediction of ridership using a ‘less than trainload’ sample would be insufficient on which to make multibillion dollar decisions.” This is not serious analysis; this is the same criticism that led people to disbelieve that George Gallup could forecast elections by polling just a few thousand voters. The relevant paragraph from the ridership model that they could does mention that 400 riders means they results are “less precise than the reported analysis indicates,” but the same passage says later, which they do not quote, that the problem comes from having polled only 51 air travelers, where they would like 150-200 people per mode. Fortunately they polled 300 drivers, and it is auto/rail mode split forecast that is hard, while air/rail is a fairly straightforward function of travel time – see figure 1 of an EU air/rail report.

Now, in lieu of the ridership model that the report criticizes, it offers sanity checks. These are normally a useful check on wildly inaccurate estimates, and if done in the 1990s would have made it clear Taiwan was not going to have 180,000 riders a day, and even its present-day traffic of 110,000 is a miracle. Cox and Moore offer two sanity checks. First is the aforementioned comparison to car and airplane travel time; that one can be disposed of due to fraudulent numbers. Another is a comparison to the Acela between New York and Washington. If the Acela only gets 2 million riders per year, they argue on page 35, how can Victorville-Las Vegas get 9 million?

Of course, people who have taken Amtrak know that the Acela is only about one-third of the ridership on the Northeast Corridor, and the time travel difference between Acela and Regional trains is small enough that the distinction is one of branding and service class. Amtrak claims on PDF-page 41 of its Northeast Corridor Master Plan that 70% of the corridor’s riders (of whom there are 11 million) are on the New York-Washington segment, so that’s already nearly 8 million, not 2 million. Further, the Acela is a slow train – its average speed, 130 km/h south of New York, is not much better than that of the legacy express trains that the TGV replaced; the average speed of the Regional is worse. To argue that XpressWest is just like Acela, Cox and Moore do not offer a serious model of the effect of access and egress times on ridership, but instead issue platitudes about a train that stops 40 miles outside the city.

To see how professionals model ridership, see for example Reinhard Clever’s thesis (the relevant pages are 26-33) as well as a short note of his regarding last-mile connectivity. Transfers, he argues, are less onerous at the origin end of the trip than at the destination end: if they must transfer, 55% of riders prefer to do so at the origin end, 22% in the middle, and 22% at the end. Likewise, commuters in auto-oriented suburbs of transit cities (the example given is Toronto) drive long distances to park-and-rides, but balk at transferring from the city-center station to the subway. Normally the origin end is likely to be the smaller city, but in the case of XpressWest, Las Vegas is the destination rather than the origin. As a result, it is unrealistic to expect significant ridership from Las Vegas residents traveling to Los Angeles (and XpressWest is not assuming any), but quite realistic to expect riders to go in the opposite direction.

Finally, the cost overrun projection is fraudulent. As Cox did in the report about Florida, on PDF-page 40 he is comparing a simple line in a freeway median to the Central Valley segment of California HSR, a line with substantial viaducts and grade separations. To his credit, he no longer includes the 11-point rubric of his Florida report, which overemphasized relatively small components of the cost like electrification and underemphasized civil infrastructure. Instead, the report just says it’s unrealistic to expect cost to be lower than in the Central Valley, without further explanation except that the Central Valley is flat; the need for plenty of grade separations and viaducts is not mentioned.

This could be attributed to a simple mistake, but in fact footnote 76 argues based on the simplicity of the terrain and the ample space in the median that widening I-15 will be cheap, only $1.6-2.5 million per lane-km ($2.6-3.9 million per lane-mile) in both directions. No connection is made with the fact that a grade-separated median is not available to California HSR. In fact California is planning to widen Route 99 from 4 lanes to 6 at $6 billion (PDF p. 22); it is unclear to me how long of a stretch of 99 is under consideration, but the full length including segments north of Sacramento is 640 km, of which about 240 appears to be already 6-lane, which would make the cost $15 $7.5 million (it would include freeway conversion, but the same issue with grade separations is true of California HSR and has been the primary driver of cost overruns in the Central Valley). The construction cost difference between the Central Valley and XpressWest is a factor of 2; perhaps it’s Cox and Moore who, in assuming one ninth to one sixth one fifth to one third the per-km cost of CA 99’s Interstate conversion, are lowballing costs for their own favored project, and not XpressWest. (Update: I misread the footnote, and the cost contained therein is $1.6-2.5 million per unidirectional lane-km.)

No other argument is presented that costs will run over, except that according to Flyvbjerg they might. Since the projected costs are well within California’s per-km cost if one omits the viaducts, tunnels, and grade separations, we can assume that costs are likely to stay under control. In fact the cost escalations on international HSR lines have typically come from heavy tunneling, which is less predictable than at-grade construction. The at-grade lines in France have stayed within budget. In Norway the 50% cost overrun of the airport train was centered on a difficult tunnel. German lines run over too, but have significant tunneling as well, and the recent overruns in Korea (subtracting the first phase, comparing cost projections from 2010 and 2000 shows a 40% overrun) were in the nearly-50%-in-tunnel second phase. But in Japan, as far as I can tell recent Shinkansen construction is on-budget despite heavy tunneling, and the same is true of AVE construction in Spain. Tunnels, we can conclude, are riskier than at-grade construction; in fact the biggest risk for at-grade construction, as seen in the California HSR project, is that viaducts or tunnels will be needed due to further engineering or environmental work, and running alongside a freeway minimizes the chance.

Because the study’s attempts to model cost and ridership are so weak, it should not be considered a serious challenge to XpressWest. Cox has had a troubled relationship with the truth in the past, and there is no argument he won’t make, no matter how ridiculous, to argue for the superiority of car travel over rail and mass transit. It’s actually the strong arguments that he fails to make – for example, regarding a possible comparison between Las Vegas and overheated East Asian Tiger economies. (For the record, I think Las Vegas is going to come out solid in such comparison.)

It is in reality quite easy for HSR to make enough money to cover above-the-rail expenses, and even track maintenance is quite cheap at about $125,000 per double track-km, but covering interest expenses is harder. Despite the canard that only the LGV Sud-Est and the Tokaido Shinkansen have paid back their interest, sourced to as far as I can tell just one person and reproduced by Cox and Moore on PDF-page 43, in reality multiple intercity railroads are profitable even including interests. This includes all three main island Shinkansen operators in Japan, SNCF, and DB. The belief that they are not comes from two sources: in Europe, conflation of subsidized commuter lines with profitable intercity lines, which are usually run by the same national railroads, and in Japan, the fact that the government wiped the accumulated operating deficit debt of Japan National Railways after splitting and privatizing it, but not Shinkansen construction debt (see references here).

So if Reason is so wrong, and XpressWest will likely meet both ridership and cost projections, what are my problems? In one word: uncertainty. Projected XpressWest revenue, on PDF-page 54 of the ridership model, is about $500 million per year in today’s money. Long-term inflation-protected federal debt is unusually cheap right now and this could make XpressWest a prudent investment – as of the time of this writing, the US can sell 30-year inflation-protected bonds at an interest rate of 0.5%, or $32 million on a $6.5 billion loan. HSR margins in Europe are low, but in Taiwan the margin in 2009, excluding interest, was 25%, which is enough (that said, despite falling far short of expectations, Taiwan HSR has very high ridership for what it is, and of course lower ridership means lower margins independently of interest rates).

But 0.5% interest is for safe investments, and infrastructure is not a safe investment. The claims that costs would run over and ridership would fall short are probably going to be proven wrong if construction goes through, making the project a success, though not a smashing success. But if the reduction in Las Vegas’s growth proves permanent and not just one recession, or if casino gambling declines, or if station access time proves more important than previously assumed in the model, or one of many other things that could go wrong, operating profits will decline.

This is what Cox fails to understand when quoting Flyvbjerg. Flyvbjerg talks about an average cost overrun – but more than that, he is concerned with risk. Many projects stay within budget or run over just a little, but a few cost several times as much as the original estimate. Telling the Big Digs and East Side Accesses apart from the Madrid Metro extensions is hard, and this is why it’s not appropriate to compute interest rates based on the borrowing costs available to the federal government.

At a riskier rate of return, things are troubling, as Paul Druce notes: he compared revenue estimates to the 30-year T-bill interest rates as of last year (3.75%), and found that operating margins would need to be above 25% until 2031 to maintain profitability. XpressWest is now looking for a larger loan than Paul assumed, but at a real rate of return of 2 or 3%, interest would indeed bite into the cost. If the project is that risky, it should therefore not be funded. That said, European transit projects tend to go ahead with a benefit-cost ratio higher than 1.2, which is certainly true of this project.

So the question is twofold. First, whether it’s sensible to lock in low interest rates and fund projects that would not be able to pay back their loans at the interest rates of a fast-growing economy. Second, how risky the project is. The first question is easier: on a pure cost-benefit analysis, the federal government can afford to lose a few billion dollars on a small number of bad investments, as long as it makes it up with enough successes, and this makes the net financial cost of the project to the government low (but positive, since it bears downside risk but does not benefit from the upside except indirectly through taxes); on top of this, precisely because the High Desert and Nevada are in deep recession, this project has additional economic benefit. The recession won’t last forever, but it exists now and will probably continue for the duration of construction.

I believe the answer to the second question is that it’s of moderate to high risk. The risk of cost escalations is low because the right-of-way is already secured and there is no difficult civil infrastructure. The risk of ridership shortfalls is more substantial – ridership estimates, especially of road/rail mode shares, have an inherent uncertainty, and on top of that the recession could cause permanent damage to Las Vegas. In addition, the strong Friday peak of travel to Las Vegas means that more rolling stock and station infrastructure will be needed relative to ridership than elsewhere, driving down operating margins.

The most troubling part of the project is that growing ridership will require a connection to Los Angeles, and because it requires a difficult mountain crossing, XpressWest is not interested in paying for it. Its current plan is to wait until California HSR opens to the LA Basin, and then link up with a line from Victorville to Palmdale. This is the real cost risk, and not the notion that at-grade rail construction is going to present the same difficulties as urban viaducts and mountain tunnels. In particular, California HSR will need to reconsider how to get from the Central Valley to Los Angeles, and the alternative that links with XpressWest goes through Palmdale, which appears to be more expensive by a few billion dollars than a straighter route through the Grapevine and Tejon Pass.

Since there is no hope for fast enough recovery that interest rates will rise, forcing early investment, it’s fine to wait. I would seriously suggest that the FRA delay decision until after the election, and if the Democrats win control of both the White House and Congress, wait a few more months until there is or is not a federal bill to fund HSR. The important thing to do is avoid biasing California toward an alternative that costs it several billion more dollars for the benefits of the XpressWest operation. Although California seems set on Palmdale, it is feasible that the amount of money Congress will make available for it in six months is enough for an initial operating segment if and only if it switches to the cheaper Grapevine alignment, and then the plan should be to try connecting XpressWest to the LA Basin much later, through tunnels through Cajon Pass. (In fact, if there is any way to get a cost estimate quickly, I would propose that, to see if it’s a reasonable alternative to Palmdale.)

If it’s a yes or no decision then I’m leaning toward yes, but not at any cost. If there is serious competition for other rail projects with higher or less risky benefits, then they should be funded ahead of XpressWest. If the decision biases California against the Grapevine, and the amount of funding available to it (from a separate pot of money, as it’s not asking for an FRA loan) is such that Palmdale would force unconscionable compromises elsewhere, then to protect the more important California HSR project XpressWest should be delayed even at the cost of potentially missing the window in which it can be funded.

But despite my doubts, it’s not a high-speed train to nowhere. It’s a high-speed train from the edge of a large metro area to a major leisure travel destination, and the cost of borrowing is so low that the federal government can expect to make its money back in ordinary circumstances. There is enough cushion against a ridership shortfall that the ordinary uncertainties expected are a small deal, and although a very large shortfall is likelier than for, say, the Northeast Corridor, it’s not probable enough to warrant denying a loan application. If Reason succeeds in canceling the line, it will join Florida HSR as a line that could have had great promise but succumbed to lobbying and fraud.