The Case for Expected RevPAR

In the late 1970s, a night watchman at a pork-and-beans cannery named Bill James started asking a simple question about baseball: what if the stats everyone relied on were measuring the wrong thing? Batting average, it turned out, was a mediocre predictor of how many runs a player actually helped produce. On-base percentage (crediting a walk the same as a single) was a much better predictor. James self-published his findings in an annual pamphlet, without a lot of coverage. Two decades later, Billy Beane used the same idea to build a playoff-caliber Oakland A’s roster on baseball’s smallest budget. Michael Lewis wrote it up as Moneyball, by 2002 the Red Sox had hired James himself and only two years later ended the Curse of the Bambino by winning the World Series doing it his way, going on to win again in 2007 and 2013 with James as Senior Advisor. Every front office in the sport now runs on a version of James’s insight.

Hospitality had almost exactly the same moment, at almost exactly the same time — and then stopped.

Hotels had their on-base percentage. They just never developed a sequel.

Before the late 1980s, hotel performance was judged mostly on occupancy: rooms filled, night after night, the equivalent of judging a hitter by batting average alone. Eric Orkin’s 1988 Cornell Quarterly article made the case that occupancy or rate in isolation missed the real signal, and Sheryl Kimes formally proposed combining the two into a single number the following year (Hospitality Net). STR standardized it into an industry-wide benchmark soon after. RevPAR (Revenue Per Available Room) did for hotels what OBP did for hitters, combining pricing power and volume into one metric that captured value the old stat couldn’t see. It was a genuine, industry-reshaping upgrade, and it is now roughly 35 years old (CoStar).

Baseball didn’t stop at on-base percentage. Statistical refinement kept going, eventually to Wins Above Replacement (WAR), a single metric that tries to isolate a player’s actual value by adjusting for context a raw stat can’t see: the ballpark, the position, the season. WAR exists because on-base percentage, useful as it was, still couldn’t tell you whether a shortstop’s .380 OBP in Coors Field in 1999 meant the same thing as a first baseman’s .380 OBP in Dodger Stadium in 1968. The hospitality industry never built its version of that question. RevPAR still can’t tell you whether a hotel’s number this month reflects durable market strength or a single date on a calendar.

This isn’t just a measurement quibble.

Trailing 12-month RevPAR is a standard underwriting input for refinancings, for debt-service-coverage covenants, and for the appraisal a lender leans on to decide how much a hotel is worth. A property that closes out a year with an artificially inflated RevPAR looks, on paper, like a stronger credit than it may be, right up until the number reverts and the loan turns out to be sized against a month that will never come back. Corporate finance solved a close cousin of this problem decades ago, and nobody blinks at it anymore. Adjusted EBITDA exists precisely because lenders and buyers got tired of raw earnings distorted by a one-time litigation settlement, a restructuring charge, or an extraordinary gain, so accountants built a standard practice of stripping those items out to show the sustainable earning power underneath. It’s now the number M&A deals and credit agreements are actually built on — nobody treats it as a gimmick. Real estate has never built its equivalent for a market’s revenue metric, which means the same distortion that finance normalized out of a company’s income statement 30 years ago is still sitting untouched in every hotel’s top line.

This blind spot is not hypothetical.

Washington, D.C. hotel rates jumped 137% year over year for this year’s Fourth of July, tied to the country’s 250th anniversary (Plus500); Philadelphia and Boston moved similarly, partly on World Cup matches (Reuters). That number will sit in trailing RevPAR next spring exactly as if it were durable growth, and the industry already has decades of evidence for what tends to happen next. Beijing’s RevPAR grew roughly 430% during the 2008 Olympics, then fell for two years as the rooms built for the Games sat empty (JLL). Rio’s RevPAR grew 278% during the 2016 Games; within a year occupancy had collapsed from the mid-90s to the high-30s, sixteen hotels closed, and the market spent three straight years in decline (Hospitality Net; Brazil Reports). Qatar’s Hotel Performance Index spiked to 314 during the 2022 World Cup and “normalized close to pre-event levels” within twelve months (KPMG). Atlanta and Salt Lake City each saw double-digit RevPAR declines the year immediately following their Olympics (NEREJ). London 2012 is the exception that proves the point: RevPAR kept growing the year after, because London didn’t overbuild the way Beijing and Athens did (JLL). Context (how a market absorbs the spike, not just the size of it) is the whole question, and raw RevPAR has no way to answer it.

This isn’t just a thought experiment. This is already happening in other segments of real estate, as well as in other industries that faced the exact same problem. Real estate valuation has its own early version: CAPE Analytics recently launched what it calls a first-of-its-kind predictive metric, a Liquidity Score that uses aerial imagery and machine learning across 40-plus property attributes to estimate how easily a home could sell, without waiting months for comparable transactions to catch up (CAPE Analytics). Sports has continued to go further and faster. Soccer’s Expected Goals model, trained on nearly a million historical shots, replaced “shots on target” with a number that captures how good a scoring chance actually was, and now anchors a whole family of derivative stats (BBC). The NFL’s Next Gen Stats program has gone from one machine-learning model in 2018 to more than 75 running today, generating metrics, like a defensive coverage stat built this year on the same architecture behind large language models, that simply didn’t exist before, because the underlying signal was invisible to a box score (SiliconANGLE). Just like in the physical world, AI enables us to move from historical metrics to a model that uses historical metrics to create actionable “Expected RevPAR”.

Moving from Historical to Expected RevPAR. STR, JLL, and others already have all the relevant data — which they will provide as a research note, market by market, one to two years after the fact. That’s fine for understanding what happened to Rio. It’s useless for underwriting what’s about to happen to Philadelphia. What’s new is the ability to run that same judgment continuously, at the level of every asset in a portfolio, not just the handful large enough to justify an analyst’s time: a model trained on the accumulated library of mega-events (from the Olympics, to the World Cup, national anniversaries, conventions and so on) that nets the known event effect out of a market’s trailing revenue the way WAR nets a hitter’s park out of his slugging line, and separately flags whether the market looks like it’s absorbing the spike like London or overbuilding for it like Beijing. The goal isn’t to forecast next month’s RevPAR. The goal is to estimate how much of today’s RevPAR reflects durable market fundamentals versus temporary calendar effects. Two numbers where there is currently one: RevPAR and an “Expected RevPAR” that tries to say what the market actually is, not just what the calendar made it look like for a week.

It took more than a century to go from Henry Chadwick’s “box score” to Bill James’s Sabermetrics, and then 25 years for James’s pamphlet to every front office using sophisticated stats reflexively. Hospitality got its OBP in 1989, but never developed its WAR. The tools to build it now exist, and other corners of real estate and other industries are already proving the pattern works. So who will build the hospitality version first, and will the rest of the industry trust the number enough to use it?