This question touches two large literatures. Neither is new, and the project borrows from both.

Agglomeration and clusters

The economic case for firms locating near each other goes back to Marshall (1890): shared labor pools, specialized suppliers and knowledge spillovers. Porter's work on clusters (1998) made the idea a management staple. Empirical work finds real effects in some settings. Greenstone, Hornbeck and Moretti (2010) show that incumbent plants' productivity rose after a large new plant opened nearby. Henderson (2003) finds scale externalities within industries for high-tech but not for capital-goods plants. Glaeser, Kallal, Scheinkman and Shleifer (1992) found that cities grew more with a diverse industrial base than with specialization. Studies of cluster size and firm performance include Folta, Cooper and Baik (2006) and Kukalis (2010). Engelberg, Ozoguz and Wang (2018) examine information spillovers inside industry clusters and market efficiency.

Geography and stock returns

Pirinsky and Wang (2006) showed that stocks of companies headquartered in the same metropolitan area move together beyond what industry and fundamentals explain. Hong, Kubik and Stein (2008) tied local stock prices to local investor demand. García and Norli (2012) found that firms operating in only a few states earn higher stock returns (about 70 basis points a month) than geographically dispersed firms, measuring operating footprint from the states named in annual reports. Tuzel and Zhang (2017) argued that local economic conditions create local risk factors priced in returns. Parsons, Sabbatucci and Titman (2020) documented lead-lag return effects between firms headquartered in the same place but in different sectors.

Is that spread just luck?

In plain termsRandom groups of companies also produce a spread of winners and losers. Compare the real cities with random draws and judge for yourself whether the real pattern stands out.

Bay Area companies beat the whole-universe benchmark by about 5.6% a year equal-weighted and 9.2 cap-weighted, from 2012 to 2026. That is better than essentially every random group of the same size we drew (the 100th percentile). Three checks matter:

• Not only a few stars. Removing the top five contributors (Apple, Nvidia, Meta and the two Google share classes) leaves +5.1 equal-weighted and +5.9 cap-weighted. The median Bay Area stock earned 9.2% a year versus 8.0% for the universe.
• Not simply "tech". Bay Area tech beats tech elsewhere; Boston tech (−1.0 equal-weighted) and New York tech (−5.7) do not.
• Concentrated. At the county level the edge is Santa Clara County (+12.1 equal-weighted, +12.5 cap-weighted), and the Bay Area's edge over other tech is larger in 2019-2026 than in 2012-2018.

The honest limits: it is one place, chosen after looking at 35, and delisted non-S&P companies lack prices. A stress test shows the equal-weighted edge would vanish only if about 10% to 17% of firms were missing and had collapsed, against roughly 5% missing in both the Bay Area and the whole universe.

An investor can use a city as a warning as well as a recommendation, so the weak side matters as much as the strong side. We compared the real cities with random groups of companies of the same sizes (each company keeping one random label for the whole period), repeated 150 times.

What this says:

• Most cities trail the market, and that is normal. A few large winners pull the market average up, so most baskets trail it. Random groups trail in both ways about 15 times out of 35, close to the real 18.
• The weak tail is somewhat heavier than chance, but not conclusively. Six cities beat by chance only about 7% of the time. The distribution of results across cities is a little lumpier than random, mostly on the downside, and not by enough to rule out luck.
• The worst cities were Denver (−10.5% a year), Salt Lake City (−9.1), Philadelphia (−7.2), Kansas City (−7.2), Las Vegas (−6.7) and Tampa (−6.7). Denver is notable on its own, ranking below every random group of its size. But picking the worst of 35 cities exaggerates how bad it is, and chance alone often produces a worst city that bad.
• Lagging did not repeat across the two halves of the period. A city's rank in 2012-2018 told us essentially nothing about its rank in 2019-2026 (rank correlation 0.05).

What if "laggard" just means below average?

In plain termsUsing the looser definition, being behind the market in one year made a city slightly more likely to be behind the next year, and a few cities were behind most years. But that pattern is only a little stronger than random groups produce, and it did not carry across the two halves of the period.

• Some persistence from year to year. A city that lagged was somewhat more likely to lag again, a small step beyond chance.
• Most often behind: Tampa (behind in 87% of years), Denver and Salt Lake City (80%), then Philadelphia, New York and Houston (73%). The Bay Area and Indianapolis were behind in only 20% of years.
• No stable laggard label. Cities that were behind in 2012-2018 were, more often than chance, ahead in 2019-2026. That reversal could reflect the market's later shift toward mega-cap technology, which favors the Bay Area and Seattle, and with only two halves it may be noise.

So the looser definition does not rescue a "bad city culture" signal, but it does fit your intuition that the Bay Area is rarely behind and that some cities are behind more often than not. Whether that reflects corporate culture or just industry mix and the technology boom is what a fundamentals test could answer.

The takeaway is that the clearest signals are the Bay Area on the upside and some unusually weak cities on the downside, but with 35 cities and noisy data, we cannot say that city-level corporate culture reliably predicts performance. What would settle it is repeating the test with more years, and checking whether lagging cities show weak sales and profits, not just weak stock returns.

Return rankings did not carry from one period to the next. Volatility rankings did: cities ranked by how much they swung in one period came out in nearly the same order in the next (a rank correlation of 0.88, where 1.0 is identical). Minneapolis-St. Paul (annualized volatility 15.5% versus 19.1% for industry-matched random baskets) and Chicago (16.2% versus 18.2%) are calmer than their industry mix explains. A 40/30/30 mix of the Bay Area, Chicago-Minneapolis and defensive industries cut the 2007-09 drawdown from about −51% for the equal-weighted market to about −44%. That mix was chosen in-sample, so treat it as an illustration, not a forecast.

In shocks, industry, not geography, did most of the protecting: staples, utilities and pharma held up in recessions and tariff shocks, and energy helped only in supply and inflation shocks (2022, Ukraine).

After removing each stock's industry movement, firms in the same city but different industries still move together slightly more than firms in different cities (permutation z = 2.3). The detail is sharper at the county level:

How to read the table: the correlation columns measure how closely monthly returns move together after removing each stock's industry. "z" says how far the observed number is from what shuffled city labels produce; above about 2 is usually taken as meaningful, and near 0 means no signal.

The effect is small in absolute terms, and it did not translate into outperformance: a metro portfolio made of its core counties performed about like the whole metro. But it is the relationship Patrick's hypothesis predicts, and it is concentrated where a "local business community" would be.

Headquarters or economy?

In plain termsThe tie shows up among companies that sell across the country, not among local banks and utilities. That points to shared executives, lenders and advisers, not the local economy.

A company's headquarters is not where its plants, stores and customers are. If shared local conditions drove the co-movement, firms that operate locally (banks, insurers, utilities, REITs) should show it most. They show none (z = −0.2). Dispersed firms (tech, pharma, industrials and others) show it clearly (z = +5.7). That favors the headquarters-community channel. The caveats are real: industry is a rough proxy for operating footprint, and the local-operating test has fewer pairs and less power.

The textbook expectation is yes. In returns, the pooled answer is close to no. Regressing each firm's return (relative to its industry that year) on the number of same-industry headquarters neighbors gives about +0.6% a year per log-peer (a measure where each doubling of neighbors counts the same), with a 95% confidence range of roughly −0.4 to +1.7: the true effect is probably in that range, which includes zero. Outside the Bay Area it is +0.05. Houston energy firms trailed energy firms elsewhere by about 4% a year, which fits Patrick's point that Houston's oil industry sits there for geographic reasons, not because proximity makes firms better.

The result is less clean than that headline, and the page should say so:

• Comparing each firm with its own past (a "firm fixed effects" test) gives +3.1% per log-peer (t = 2.8). That could be real agglomeration or clusters growing after their firms succeed.
• Cap-weighted, the effect is +2.6 (t = 2.6), driven by large Bay Area tech names.
• Tech hardware is positive (t = 2.2); industrials and insurance are negative. With 16 industry groups, some of this is chance.
• Of 38 industry-county clusters with at least six firms, 3 beat the same industry elsewhere at t above 2 (about 1 expected) and none lagged that badly. The three are Stamford, San Jose hardware and San Jose software.

The most defensible reading: no general return premium for HQ proximity to same-industry firms, with any effect concentrated in Silicon Valley technology. Even that is limited. Stock prices already reflect expected advantages, so returns detect only benefits the market did not anticipate. Whether proximity helps a firm's sales, margins or productivity is a different question, and the right test uses financial statements.

What this chart does not fix. The company list is today’s listings plus former S&P 500 members, so firms that failed or were delisted outside the S&P 500 are missing, which flatters every line (small companies most). The “all listed companies” view covers 56 regions: the 46 averaging at least 8 companies, plus ten regions with fewer companies but large market value, such as Northwest Arkansas (Walmart), Boise (Micron) and Kalamazoo (Stryker). It starts with about 1,700 firms in 2007 and grows to about 4,000, applies no size floor, and caps yearly returns at +500% and −95%, so it will not match the 2012–2026 study above. The smallest regions rest on very few companies, so one stock can move them; the table under the chart lists each region’s company count and market value.

The market-cap-weighted funds start in December 2011, because the SEC share counts that the weights need are too sparse before then. Each company is weighted by its market value at the start of the year. Share counts cover about 89% of listed companies in 2012 and 96% in 2026; companies without them are left out of the weighted funds (for example Constellation Brands, Tyson and Erie Indemnity), and gaps for S&P 500 companies were filled from the nearest reported year, adjusted for splits. Market values above $25 billion for companies that were never in the S&P 500 were dropped as data errors unless checked by hand. Giants dominate: a region with one very large company is mostly that company. In the S&P 500 view, 95% of company-years have share data, and Dublin has no market-cap fund because its foreign filers lack the share counts. The S&P 500 view uses point-in-time membership and headquarters back to 2006, but roughly half the companies that left the index have no price history; a few small cities rest on two to seven companies, so one stock moves them. Year-end values only.