A reminder on the data, as in earlier instalments: the figures come from the Historical Housing Prices project — 2.7 million newspaper listings across 30 cities (Lyons, Shertzer, Gray & Agorastos, QJE 2025; NBER w32593, VoxEU). This is also the period where the new data earn their keep most clearly. City-level sale prices don’t exist in the standard sources before 1975 (when the FHFA indices begin), and city-level market rents don’t exist anywhere before the 1990s. So almost everything below is something you simply cannot see in the data economists have used until now.

51. Real house prices did rise after the war — the old index missed it

The long-relied-upon Shiller index shows real house prices going more or less sideways from 1950 to 1995. The HHP data disagree: national real prices rose about 21% between 1948 and 1970, and kept climbing thereafter. American housing started getting more expensive, in real terms, decades earlier than the standard story allows.

52. The national calm hid a widening gap between cities

That modest national rise was an average of wildly different fates. Between 1948 and 1970, real sale prices ranged from +50% in Boston and +43% in Los Angeles down to −11% in St Louis and −12% in Memphis. Same country, same era, opposite directions.

53. California pulled away first — and early

We tend to date California’s housing exceptionalism to the 1980s. The data put it a generation earlier. Los Angeles real prices were already 50% above their 1948 level by 1960; San Diego and San Francisco were 30%+ above by 1970. The coastal premium was visible by 1960 — fifteen years before the city indices everyone uses even begin.

54. The Rust Belt’s housing weakness began in the 1950s

The mirror image of California is the industrial heartland. St Louis real prices fell about 11% between 1948 and 1970 and stayed below their 1948 level right through 1975; Cleveland, Cincinnati, Pittsburgh and Detroit were flat or falling. The Rust Belt’s housing story is usually told as a post-1970s, post-deindustrialization collapse. In fact its homes were losing real value through the supposed golden age.

55. This split tracked building, not just demand

Why did some cities soar and others stall? Largely supply. Cities that entered the postwar period with ample housing (Philadelphia, St Louis) or that kept building freely (Atlanta) stayed cheap. Cities that slowed construction — Los Angeles and San Francisco above all — saw prices rise. The zoning-and-supply debates that erupted in the 1970s were describing a divergence that had already begun.

56. The postwar rent spike hit some cities three times harder than others

Here is something only market-rent data can show. As wartime rent controls lapsed in the mid-to-late 1940s, real market rents surged across the country — but by wildly different amounts. Miami’s real rents peaked 227% above their 1941 level in 1947; Los Angeles peaked +140% in 1946; Tampa +115%; Phoenix +114%. At the other extreme, some cities barely felt it at all: Baltimore’s real rents rose just 1%, and Charleston’s kept falling right through the war, 27% below their 1941 level by 1944. Nationally, the spike peaked at +35% above 1941, in 1948 — but that single number hides a country experiencing completely different housing markets at the same time.

57. The spike was temporary almost everywhere — except where cities couldn’t build

Nationally, the postwar rent surge fully unwound: by 1960, real rents were only 4% above their 1941 level — the whole spike had given itself back. Most cities followed the same path within a decade of their peak: Phoenix was back to normal by 1951 (7 years), St Louis by 1957 (9 years) — the same cities, from fact 55, that kept building through the period. But the cities that couldn’t build never gave the spike back. Los Angeles’s real rents were still roughly 60% above their 1941 level in 1971, and Tampa’s premium had actually grown — to 126% above 1941, bigger than its original 1947 peak. The same supply constraint that was pulling sale prices apart (fact 55) was doing the same thing to rents: a temporary wartime shock in cities that could build, a permanent one in cities that couldn’t.

58. Prices and rents permanently parted ways

Once the postwar rent spike fades from the picture (facts 56–57), a cleaner pattern emerges: nationally, real sale prices kept climbing through the postwar decades and beyond (fact 61), while real rents settled and stayed roughly flat from the early 1950s on. That is the start of the permanent divergence between sale prices and rents that defines modern housing markets. Only a dataset with both segments, city by city, and long enough to look past a wartime rent shock, can show cleanly when and where this began.

59. Yields collapsed as the modern mortgage arrived

If prices rise while rents don’t, the rental yield falls — and it did, fast. The national gross rent-to-price ratio fell from about 10.6% in 1948 to 7.4% by 1960, the sharpest compression in the whole record. The cause was financial: the long, fully-amortizing, high-loan-to-value mortgage spread between the 1930s and 1960s, steadily cutting the cost of owning relative to renting and letting buyers bid prices up against the same rents.

60. Yield compression had two opposite faces

The same falling yield meant different things in different places. In California, yields fell because prices raced ahead of rents. In the industrial South and Midwest, yields fell because rents collapsed faster than prices — the rent-to-price ratio dropped ~45% in Louisville and ~41% in New Orleans and St Louis, driven by falling rents, not rising prices. One statistic, two completely different stories.

61. The calm of 1948–1970 was the leading edge of the longest boom in the whole record

Zoom out from any single window and something remarkable appears. From its 1949 trough, the national real sale-price index rose 77% over the next thirty years to a peak in 1979 — without a single interruption worse than about 6%. (The two dents: −6.4% in 1949 itself and −5.3% in 1958; every other year in the run was a new real high.) That makes it roughly two and a half times longer than the next-longest expansion in the entire 1890–2006 record — the 1994–2006 bubble, twelve years, +76% — and about four times the length of a typical prewar cycle (5 to 8 years, peak to peak: +23% in 1900–08, +62% in 1919–27, +46% in the immediate 1942–47 postwar surge itself). What looked, from inside 1948–1970, like modest, steady growth was in fact the first two-thirds of the longest unbroken run American housing has ever had.

One more thing happened alongside all this — worth knowing, even though it isn’t something HHP alone can claim. The national homeownership rate rose a record ~13.7 points in the 1950s alone, reaching nearly 62% by 1960, up from a Depression-era 38%. That number comes from the US Census, not from HHP; it’s useful context for everything above — the mortgage revolution behind fact 59, the price boom behind fact 61 — but it isn’t itself a discovery this dataset can take credit for.

Next time (Part VII): the Great Inflation of 1968–1983 — including 1979–82, the moment the thirty-year calm identified in fact 61 finally breaks — and how badly the official statistics mismeasured what was happening to both rents and prices along the way.

A reminder on the data, as in earlier instalments: the figures come from the Historical Housing Prices project — 2.7 million newspaper listings across 30 cities (Lyons, Shertzer, Gray & Agorastos, QJE 2026; NBER w32593, VoxEU). Two things make them especially useful here. First, they are annual and city-level, so we can watch the Depression arrive in different cities at different times and with very different force. Second, they measure sales and rental prices separately. City changes are nominal unless “real” is stated; “real” means adjusted for consumer prices. (Homeownership rates are from the US Census.)

37. The housing bust led — and lagged — the stock market crash

Wall Street peaked in September 1929 and crashed the following month — the stock market then kept sliding for almost three years, bottoming in July 1932 some 89% below its peak. Housing ran on a slower clock. Prices peaked earlier, in 1928, and adjusting for inflation they fell by one quarter, not bottoming out until 1933. Some individual cities did not hit bottom until 1935 — well after stocks had already turned.

Housing is slow-moving and illiquid; the air came out of the housing bubble gradually, not in one season.

Subscribe now

39. Florida was closer to cause than effect

The decade’s most spectacular housing market collapses were closer to the cause of the Great Depression than cities swept up in its aftermath. The Florida land bubble of the early 1920s burst early: Miami fell 82% from its 1925 peak and Tampa 72%, both bottoming before the worst of the national downturn. By the time the rest of the country was falling, Florida was already on the floor.

38. Detroit had the deepest fall of any city — down almost three quarters

Setting aside the two Florida cities, by far the most affected market was the Motor City. From its 1926 peak to its 1933 trough, Detroit sale prices fell 73% in nominal terms (over 60% in real terms). Detroit had also been the most expensive market in the country relative to its later self in 1926 — the 1910s car boom had pushed it highest, but this just meant it had the furthest to fall.

40. The hardest-hit cities were industrial or over-built

After Detroit and the two Florida cities, the deepest falls clustered in the industrial Midwest and the South and the speculative West: in Atlanta, Dallas, New Orleans and Phoenix, prices fell by more than a half — and so too in Los Angeles, Portland and Seattle. Cleveland, Chicago and Philadelphia also saw falls of more than one half. Cities that had built and borrowed hardest in the 1920s paid the most in the 1930s.

It’s worth noting that it’s not quite as neat as that: Cincinnati and Nashville were among the cities with the smallest falls.

41. Prices fell least in Washington, D.C.

At the other extreme sat the federal city. D.C. prices fell least of all. On the peak-to-trough basis in the chart above — its 1924 high to its 1935 low — Washington lost about 30%, the smallest drop of any city in the data. And measured over the core Depression window of 1928 to 1933, the fall was gentler still: around 20% in nominal terms. Indeed, in that core five-year window, consumer prices fell even faster, meaning in real terms, housing prices in DC were about 5% higher at the end of the Depression than at its start.

As the New Deal expanded the federal government, Washington's economy — and its housing market — was insulated from the collapse engulfing everywhere else. It was a company town, and the company was the U.S. government.

42. This was a bust in sales prices, not in rents

Market rents fell far less than sale prices, and in real terms they often rose. Adjusted for the deflation of the early 1930s, real rents increased in many cities through the Depression — with Washington again leading the way (up 24%) but rents in real terms also rose, by about 10%, in Charleston, Salt Lake City, Minneapolis, Boston and San Diego.

People still needed somewhere to live — and as long as they had jobs and incomes, that was reflected in rents. What collapsed was the willingness to capitalize that shelter into a purchase price. Mechanically, that means rental yields rose during the crash — the opposite of the yield compression you see before a classic asset bubble. The Depression was a financial and expectational bust, not a collapse in the value of housing. (You can see this in the chart at the top of the post: the rent line sits above the price line right through the 1930s.)

43. Where rents did fall, it tracked unemployment

The cities where rents fell, and fell hardest, lent credence to the idea that rental markets reflected the real economy, while sales markets reflected the financial one. Just as the markets where rents rose were those most insulated from job losses, the worst-affected rental markets were those where employment collapsed: Detroit (auto) and Seattle (shipping and lumber) saw rents fall by about one half. In Chicago and Pittsburgh, industrial employment fell sharply — and so did rents (by between 40% and 50%).

Where the jobs went, in general so did the rents.

Share

44. Homeownership fell in the 1930s — the only time before the 2000s

The 1920s had been an ownership boom: the national homeownership rate climbed to 43.5% by 1930, up more than 8 points over the decade. The Depression reversed it. Ownership fell 5.6 points in the 1930s — to under 38% by 1940 — the only decade-long decline in the rate before the 2007–11 bust. Foreclosure and forced sale did their work.

45. The recovery was deeply uneven — the old Northeast missed it

Recovery, when it came after 1933, did not lift all boats. Strikingly, the old industrial Northeast kept falling even as the economy turned: from 1933 to 1941, New York prices fell another 17%, St. Louis 14%, Philadelphia 7% and Boston 6%. Cities that had entered the 1930s with ample housing had no reason to build or to bid prices back up.

46. Meanwhile the Sunbelt and West began to rise

While the Northeast stalled, a different map was emerging. From 1933 to 1941 prices rose most strongly in Charleston (+74%) and Miami (+64%). In Detroit (+52%, an auto-led rebound) and Salt Lake City (+49%) they grew by half — and by almost as much in Tampa (+46%) and San Diego (+45%). In Houston and Los Angeles, they grew by about one third. A decade before the postwar Sunbelt shift became conventional wisdom, you can already see it in the housing data.

47. Rents recovered faster and more broadly than prices

Renters felt the recovery before buyers did. From 1933 to 1941, rents rose in almost every city and by more than prices — Charleston +84%, Seattle +83%, Detroit +80%, Baltimore +57%, Cleveland +51%, Louisville +50%. Demand for shelter returned well ahead of the confidence to buy it.

48. World War II set off the biggest housing boom yet

If the 1920s boom was big, the wartime boom was bigger. Between 1940 and 1948 national real sale prices rose about 40% — more than over the entire Roaring Twenties. In real, inflation-adjusted terms, Chicago rose 85%, Los Angeles 75%, Miami 67%, Dallas 62%, San Diego 59% and Tampa 56%. War production drew workers into cities far faster than housing could be built for them.

49. Market rents soared during the war — despite rent control

The federal government imposed rent control through the Office of Price Administration in 1942. And yet the HHP market rents — the rents asked of households signing a new lease — rose dramatically: almost trebling in nominal terms in many cities, including Dallas, Miami, Tampa and St. Louis. Adjusting for inflation, rents were 40–70% higher after the war across the South and West.

The likely reason is exactly the distinction at the heart of the project: controls bit hardest on contract rents (what sitting tenants paid), while market rents — and the pressure behind them — kept climbing. The gap between the two is itself a measure of how binding wartime housing shortages were. (A methodological note: HHP measures listed, new-lease rents, which can diverge from controlled contract rents and from transacted prices. Allison, Rowena and I have a new paper that looks at this in more detail.)

50. In 1948, the West was above 1928 levels, the industrial core lagged — and ownership America was born

Step back and take stock of the whole cycle. Two decades of crash, recovery and war boom left national real prices in 1948 almost exactly where they had been in 1928 — up just 2%, a twenty-year round trip. But that flat average hid a continental divide. In real terms the Pacific West and Florida had pushed well past their 1928 level — Miami +83%, San Diego +48%, Memphis +42%, with Portland, Seattle and Salt Lake City all near +40% — while the old industrial Northeast and Midwest were still below 1928: New York −16%, Philadelphia −15%, St. Louis −23%, and Detroit a full third lower (−33%), the only one of the thirty cities worth less in 1948 than 1928 even in cash terms. (One caveat: Miami and Tampa had peaked in the 1925 land bubble and Detroit in 1926, so for those three 1928 is already a little off their own peak.) The Sunbelt-and-Pacific tilt visible in the 1930s recovery and the wartime boom had, over twenty years, quietly redrawn the map of American housing.

And the boom changed how Americans lived. The homeownership rate jumped +9 points in the 1940s and a record +13.7 points in the 1950s, from a Depression-era 38% to just over 60%. The mortgage revolution the paper documents — longer terms, higher loan-to-value ratios spreading between the 1930s and 1960s — turned the wartime price boom into the mass-ownership society we still live in.

Subscribe now

Next time (Part VI): the long post-war climb, the mortgage that made it possible, and the moment in the 1970s when American house prices stopped being flat and started their modern ascent.

A reminder on the data: these figures come from the Historical Housing Prices project, built from 2.7 million newspaper listings across 30 cities (Lyons, Shertzer, Gray & Agorastos, QJE 2026; NBER w32593, VoxEU). They are annual and city-level, and — crucially — they measure rents and sale prices separately.

#26. Housing prices rose only modestly between 1910 and 1916

Compared in particular to what came next, the early 1910s were—by and large—a period of stability, or maybe better described as modest upward drift, at least until the US became directly involved in World War I. As of 1916, across the country as a whole, rents were about 11% higher than in 1910. Sales prices were 8% higher.

And while each city had its own story, most cities saw either stable prices (Pittsburgh prices rose by just 1% in this period), modest declines (St Louis saw falls of 4%) or slight increases (Boston prices rose by 8%). The same pattern is there in the rental market too: Cincinnati and Dallas rents were unchanged, Minneapolis rents were up 6% while LA rents were down 5%.

#27. Detroit boomed like nowhere else in the early 1910s

The stand-out exception to this pattern of at best modestly rising prices in the 1910s is Detroit. Sales prices rose by 70% between 1910 and 1916, more than twice the next largest increase seen, which was in Miami, itself on the cusp of a housing bubble. Among established cities, only Cleveland (33%) and Chicago (22%) could point to any substantial increase in prices in this period.

What is telling is that the increase seen in sales prices was largely matched by increases in market rents. Rents rose by 64% in the same period, again dwarfing the increases seen elsewhere in the country. Sales prices rising without any matching increases in rents is a classic sign of a bubble. Where both rise together, that tells us instead that this was booming demand and, clearly, an inability of supply to keep up. This heyday of the auto industry was, of course, what gave “Motor City” its name.

#28. After a 1900s boom, the West mostly missed out in the early 1910s

In the last post, we discussed the strength of housing prices in the West—in cities like Portland, Seattle and Los Angeles—in the 1900s, the first decade we can really see trends for the region as a whole. (Series go back further for some cities, including SF and LA, but not for most cities in the region.)

But strong demand for the West appears to have eased in the 1910s. Between 1910 and 1916, prices in Portland fell by 18%, more than anywhere else, while rents in the city fell by 24%, again the largest fall seen in the dataset. Seattle’s market was only slightly less dramatic, with prices falling 16% and rents by 15%. In both San Diego and Los Angeles, also, both sale and rental prices fell in these years, albeit by 5% in the case of LA and by up to 9% (sales; 2% rental) in San Diego.

Post-earthquake San Francisco bucks the trend (prices up 3%, rents up 13%) and its outsized importance in the region during this period means it drags up the regional average. But clearly, the West was on its own path in these years.

#29. Market rents more than doubled at the end of World War I

After this period of relative calm came extraordinary change. For the US as a whole, market rents rose by 6.3% between 1916 and 1917. This was the joint-largest jump in rents going back to 1890 (matching the increase seen in 1902).

But this was only the start: in each of the following three years, the increases seen nationwide in rents set a new record: up 15% in 1918, up 28% in 1919 and up 36% in 1920. It is tempting, and likely correct, to draw a straight line between returning troops, as well as the wider dislocation caused by moving an economy to a war footing, and large increases in rents.

By 1920, for the US as a whole market rents were, in nominal terms, more than twice what they had been in 1916, increasing 113% in just four years. These changes were without precedent, at least since the spike in rents seen in cities such as New York at the end of the Civil War, half a century earlier. And, to some extent, they largely reflected changes in the wider price level. The CPI peaked in 1920, at a level 88% above its 1916 level.

It’s true that the increases in wider consumer prices in 1917 and 1918 were greater than that seen in market rents—meaning in ‘real terms’, rents were actually lower in 1918 than in 1916. But the increases seen in 1919 and 1920 in rents were much greater than those seen in the general price level: an increase of almost 75% compared to an increase in wider prices of “just” one third.

A New York policymaker referred to the housing market in these years as a ‘housing famine’ and it’s not hard to see why.

#30. Texan rents tripled in the late 1910s

It is around this national figure of rents increasing 113% that we can then assess city-level figures. Patterns in rent changes in these years are less obvious than those before or after. There are Southern and Western cities with among the largest increases (Dallas and LA, for example) and also among the smallest increases (Nashville and Salt Lake City).

Three cities saw market rents more than triple between 1916 and 1920, though: our two Texan cities, Dallas and Houston—first and third respectively with increases of 225% and 204%—and Los Angeles, where rents rose 218%. There are other Southern cities with large increases (Atlanta and New Orleans) but the presence of both Texan cities in the top three for rent increases suggests something more systematic in the local economy at this point, especially as the two cities also saw above-average increases in sales prices during the same four years.

#31. Sales prices rose 1916-1920 — but by less than rents or wider prices

Sale prices rose as well, just later and more slowly: roughly two-thirds higher in 1920 than 1916 in nominal terms. But with general prices up nearly 90%, real sale prices in 1920 were actually below their 1916 level — they did not recover in real terms until 1921.

And the sales boom, when it came, had a different geography from the rent boom. Between 1914 and 1920 it was the industrial heartland that capitalized the war economy: Cleveland prices rose 147% and Baltimore 140%, with Detroit (+94%), New York (+83%) and Chicago (+80%) close behind. Where the war rent surge was a Sunbelt-and-Pacific story, the war sales boom was a Rust-Belt one.

#32. Some cities Roared in the Twenties (some didn’t)

Consumer prices peaked in 1920 at just over twice their 1914 level. While there was a mini-reversal, with prices rising 1924-1926, that downward trend would continue all the way until 1933. By then, the general price level was 35% lower than it had been at the worst of the post-war spike. This matters when putting the performance of both sales prices and rental prices into their correct perspective for the Roaring Twenties.

In fact #31 above, I mentioned that nominal sales prices rose by an astonishing 28% in 1920 (and earlier that rents rose by 36% the same year). But the substantial inflation in the economy means the real (inflation-adjusted) changes are still large but not as eye-watering: across the US as a whole, sales prices were in real terms 10.8% higher in 1920 than a year before.

Make no mistake, that is still by far the largest increase seen across the previous three decades (taking inflation into account). But it was just the start. Because, while consumer prices started their prolonged fall in 1921, the increase in sales prices was just getting started. In nominal terms, sales prices rose by less than 5% in 1921 but — accounting for inflation — in real terms, prices rose by almost 17%. Something similar happened in 1922 while in 1923 and 1924, there were further increases.

All in, nominal sales prices rose by a further 29% between 1920 and 1926, at a time when consumer prices (and rents — see below) fell. Combined with those falls in consumer prices, in real terms sales prices were 73% higher in 1928 than a decade earlier.

But… this increase was far from even. In Salt Lake City, prices in 1926 were unchanged on 1920 — in the South, as I discuss more below, prices in many cities fell in these years. But in St Louis, which had seen very modest inflation 1916-1920, prices rose 77%. A bunch of cities — including Boston, Cincinnati, LA, Louisville and Philadelphia — saw prices increase by almost half in those few years.

#33. The Florida bubble was not characterized by a fall in housing yields

The figure above — which gives the percentage change in housing prices, by region for 1920 to 1926 — has one clear stand-out feature: Florida. As Barry Eichengreen wrote in ‘Hall of Mirrors’, the Florida land boom of the 1920s is not only fascinating on its own terms, it also has very clear lessons that could have been heeded ahead of the 2000s bubble/crash.

Florida here, in our thirty-city dataset, refers to Miami and Tampa, both of which can be measured from the early 1910s. The 1920-1926 timing, imposed from national cycles, even understates things somewhat: while Tampa peaked in 1926, Miami — which was closer to the epicenter — peaked in 1925, while the start of the cycle was arguably 1921 rather than 1920.

Looking just at 1921-1925, Miami rents rose by a factor of three. Sales prices, at least by our measure, “only” doubled in the same span. Some of this reflects timing: we are looking across the whole year, but the frenzy gained in intensity during 1925, so a monthly index would likely show late-1925 prices to be higher than earlier in the year. A second element is that it was a land boom and we are looking at housing. Even though the land was marked for housing, structures tend to stabilize prices — it is the land, not the structure, that swings in value precisely because it can’t be replaced.

Nonetheless, it is fascinating that rents rose as much as prices did in our two Florida cities in these early years of their history. Miami and Tampa ranked #1 and #2 of the 29 cities (Vegas ain’t there yet) for overall rent increase 1910-1926 but only #4 and #3 for sales price increases. Indeed, in Miami rents rose by more. Standard bubbles are best spotted by a gap between the fundamental value of property (as represented by its rental value) and the market value (the price). Yields, in other words, fall in a standard bubble. But in Florida, they did not.

Thus, the case of Florida in the 1920s — by virtue of its late date, high visibility and extreme volatility — might be offering us a window into the dynamics of housing markets at the birth of rapidly growing cities. “A bubble, Jim, but not as we know it…”

#34. The South (mostly) missed out on the Roaring Twenties

The concept (and indeed thrill) of the Roaring Twenties has survived a century, still featuring as the themes for nights out, dinner parties and more. One key contribution of the new housing prices series is to confirm that, across the country as a whole, housing values took part in the festivities. As mentioned above, in nominal terms housing prices almost doubled between 1918 and 1925. In real terms, the increase was a bit less (and longer drawn-out) but nonetheless unlike anything seen in preceding decades: up 73% in the decade to 1928.

But not all parts of the country took part in the Roaring Twenties. The graph below plots housing price indices, both sales and rent, for the Southern cities in our dataset and for the Northeastern cities. While prices in the north-east had risen by 117% between 1910 and 1925, prices in the south rose by only half that (62%).

What is noteworthy is that this does not seem to be related to fundamentals, at least not those captured by market rents. If anything, the shortage of accommodation — relative to demand — was greater in the south than in the northeast. Rents spiked at 2.6 times their 1910 level in the south, compared to just under 2.4 times their 1910 level in the north-east.

Going back to our classic version of a bubble — a growing gap between the rental value and the capital value of housing — the yield fell substantially in the north-east in the years after 1920 and, by 1925, was below the level seen in the 1910s. In the south, however, the yield was higher in 1925 than it had been in the 1910s.

#35. The boom was concentrated in the Northeast and in particular the Midwest

Setting Florida to one side for now, splitting the rest of the country into four regions reveals that it was not just the south that missed out on the Roaring Twenties, nor was the north-east the only place to take part. The final graph in this post plots sales and rent prices for the Western cities and for the Midwest sample from 1910 to 1925.

While not as dramatic as the Southern cities, nonetheless the overall increase in sales prices in the West 1910-1925 was significantly smaller than in the Midwest or the Northeast. In particular, the spike seen in other regions — even in the South — between 1918 and 1920 is not apparent in the West. Drilling into the data, this is driven by San Francisco, the most important segment of the Western market at this time. Prices increased by just 20% between 1918 and 1920, the third smallest increase across the 29 cities covered (only Nashville and Salt Lake City saw smaller increases).

Rents, however, did surge in these years meaning that — like the US south — Western yields in the mid-1920s were higher than they had been in the 1910s, not lower as one might expect in a frothy housing market. This marks a sharp contrast with the Midwest, where yields started to fall sharply after 1922 (the gap between the orange and blue lines at the top of the graph).

Subscribe now

#36. Rents fell, even as prices rose, in the early 1920s, due to new supply

While there are important differences by region in the 1910-1925 period, as outlined above, one thing is common to them all: rents started to fall after 1920, even as prices rose. In nominal terms, 1920 marked the peak — with rents stable in 1921 and then falling 12 of the next 13 years.

The falls later, as the Roaring Twenties give way to the Great Depression — as we’ll see in the next post — are sharper. But nonetheless, with the exception of 1923, rents fell, even if only modestly, throughout the 1920s. They were 7% lower than in 1920 at the end of the period here (1925).

New York is a great example (as well as the most important rental market in the country, then and now). Market rents fell by one quarter between 1920 and 1926, fueled by a building boom and glut of new rental homes that has never been matched since. In 1924, for example, over 100,000 new homes were built — about as many as the city built in the entire 1980s. The supply boom, in New York and across the country, was a consequence of shifting fundamentals and greater confidence — as well as of the acute shortages immediately after the war. It gets to the heart of the paradox of housing bubbles that the confidence in higher prices can coexist with a supply boom that is pushing down rents.

Next up, we’ll look at the period between the late 1920s and the mid-1940s, the infamous Great Depression as well as the years just before and after it. Stay tuned!

Share

I’m delighted to say that the paper is now conditionally accepted at the Quarterly Journal of Economics. In addressing the various comments that came up in peer review, we significantly improved the paper. And that involved going back to the sources and fixing perhaps 10-15 shorter runs across individual cities in key hard-to-measure periods, such as the 1890s (when listings for some cities were particularly scarce) or the 1940s (when rent controls had a similar effect).

Our revised indices are live on the Philly Fed website, for everyone to download. And I’m delighted to see that prominent Substackers like Brian are interested in our data!

With that revision process largely complete as of July, I have revised the original three posts to reflect any substantive changes (for example particular percentage point increases or falls). And today we move on. The last post looked at the 1890s and 1900s, finding, among other things, a substantial housing market bust from 1893. Today’s new post, Part IV, looks at the 1910s and the early 1920s.

I hope you enjoy reading it as much as I love swimming through the data and finding the stories. And let me know if there’s anything you’d like me to focus on in future posts.

In this post, I’m doing a deep-dive into the earliest part of the dataset, the 1890s and the early 1900s. And I’m going to start looking at the experience of individual cities, as well as regions.

#17. Sale prices fell by 40% in the South during the 1890s

To those interested in American economic development over time, the 1890s is perhaps best known for its 1893 panic, triggered by a combination of changing economic fortunes elsewhere and a collapse in wheat prices. The panic itself was marked by significant runs on the banking system, as the US grappled with a backlash to having chosen a gold-based monetary system. It was not only those involved in wheat that were affected — cotton-producing areas of the south were also affected, as Indian and Egyptian varieties gained a foothold in global markets.

The turmoil is evident in the sales price indices for five Southern cities covered in the HHP dataset from the 1890s. These are shown in the first figure in this piece, which benchmarks sale prices in Atlanta, Louisville, Dallas, Nashville and New Orleans, the five well-established Southern cities in HHP, together with a weighted average. (Memphis and Houston are also the dataset but they start later: 1895 and 1900 respectively.)

It’s important to note that, as suggested by the very fact that the index starts in 1890 and not earlier, the quantity and quality of data becomes poorer the further back in time. Some of this is related to city size: there are selections of listings with prices (in particular in the rental sector) back into the 1870s for a number of cities and indeed for New York City back into the 1830s. Some of this is related to the media industry itself — by 1900, every major newspaper in every city seems to have realized that real estate listings were perhaps the easiest way to generate advertising revenue.

But for many cities, the 1890s give at best limited numbers of listings with information such as price, location and size. For that reason, the earliest years can be somewhat volatile. You’ll notice that New Orleans — the biggest city of the four — only appears from 1893, because despite its size earlier years did not have sufficient volumes of sale listings in newspapers with enough information.

Despite those caveats, the overall trend is pretty clear. The entire decades of the 1890s was associated with generally falling sale prices in the US south, likely reflecting poor conditions in the cotton sector, on which those cities earned their living. The solid black line gives an overall average fall of 39% from 1890 to 1897, when they bottomed out.

Dallas shows the sharpest fall, with prices less than half their 1890 level in 1897. But Nashville and New Orleans — whose cotton-based port once rivaled New York as the busiest in the nation — also saw large fall. In Nashville, prices were 63% lower in 1897 than at the start of the decade, while prices in New Orleans fell 45% between 1894 and 1900.

#18. The 1890s bust in the South was more about finance than fundamentals

In general, we can think of two sets of factors that drive the sale price (or capital value) of housing. On the one hand, there are what might be termed ‘fundamentals’, factors that reflect underlying supply and demand forces and which, as a result, would be reflected in rental prices as well as sale prices. Under this heading come things like new supply (lowering sale and rental prices) or higher incomes (boosting both).

On the other hand, there are ‘asset factors’, reflecting the fact that housing is a long-lived asset. Financial considerations, including the cost of capital — not just interest rates, but the wider user or opportunity cost — will affect the ratio between sale and rental prices. While interest rates typically move within tight enough bands, and other elements of user cost such as maintenance and taxation vary even less, expectations about the future can fluctuate wildly. In the modern era, there are markets where expectations changed within the space of a few years from participants expecting prices to rise 20% or more over the next five years to expecting falls of 20% (or more).

The HHP project measures both sale and rental prices, so we can get some indicative evidence as to whether the sharp bust in sale prices in the US south in the 1890s was driven by fundamentals or by asset considerations. The second figure in this post shows the rent indices for our four southern cities and (in black) the weighted average.

As with the sales graph, 1893 is set to 1 and the axes are kept the same so that the magnitudes of changes are similar in both figures. A close inspection of the two graphs will show that the fall in sale prices outstrips the fall in rents in these cities. While sale prices fell by 40% during the 1890s in the US south, rents fell by 20%.

Note that this is not quite the same as saying fundamentals had no role. Falling rent prices are a classic symptom of declining local incomes (and indeed falling net migration or possibly net emigration from the region). The housing stock can’t move away in response to negative shocks, meaning prices (not quantities) adjust in housing downturns. Fundamentals definitely mattered — but, if the 2-to-1 ratio is roughly right, changed expectations about the future mattered just as much.

#19. Prices fell in the grain heartlands of the US in the 1890s too

The 1893 Panic was a nationwide phenomenon. As mentioned above, the wheat sector was affected as well as the cotton sector, with a collapse in wheat prices in 1893. Unsurprisingly, almost all cities in our dataset for which indices can be compiled in the 1890s show a decline in prices — even if it was short-lived.

Take the triangle (roughly) from Chicago across to Cleveland and down to Cincinnati, heavily involved in the wheat trade, indeed the heart of the so-called late 19th century “Grain Invasion” of Europe. In each of the three cities, sale prices fell during the 1890s, having peaked 1890-1892. In general, the fall in prices during the 1890s was smaller than that seen in the southern cities: about one quarter in Cincinnati and Cleveland, but over 40% in Chicago.

But these were not the largest falls outside of the south. Indeed, the largest reliably estimated fall in sale prices in the 18 cities for which indices could be calculated was in Minneapolis, further west again. There, prices are estimated to have fallen 55% between 1890 and 1899 and the 1890 peak would not be seen again until 1911. Once again, expectations rather than fundamentals appear to be driving the huge change: rents fell by ‘just’ 24% (and bottomed out earlier).

The difference is not as great for Chicago or Cincinnati. In Chicago, rents fell by 34% (compared to 42% for prices) and in Cincinnati rents fell 14% (vs 24% for prices). Again, rents bottom out first, suggesting a real economy element shifting into a financial element. (Rental data for Cleveland only start in 1894, but rents had already bottomed out by then.)

Subscribe now

#20. Housing prices in NYC mirrored those in the South in the 1890s

As the largest city in a weighted index, New York City is the most important market. This is particularly true in the 1890s, when its lead over second-largest city was greatest. (Its share of households in the 1890s was just over 30%, nearly as much as second, third and fourth combined.)

But New York, despite (or perhaps because of) its economic heft, could not escape the weak economic conditions of the decade. Both sale prices and rents in the city peaked in 1892. It would be eight years before sale prices bottomed out and when they did, they did so at a level that was 45% lower than the peak.

Rents bottomed out earlier, in 1898, at 9% lower than their pre-Panic peak. New York had done much over the preceding decades to shift its reliance from the cotton-producing areas in the south to a far larger economic hinterland stretching across the Pacific. But the Panic and the ensuing recession seems to have been deep enough to fundamentally change the relationship between sale and rental prices.

It’s worth noting that the change in sale prices in New York was more similar to the US South (30% in NYC, 40% in the US South) than in, say, Boston or Philadelphia, two other larger cities in the north-east. Prices in Boston fell by just 13% in this cycle, while those in Philly fell by 18%. This is all the more interesting given that the fall in rents in New York was smaller than the other two large cities: 14-15% in Boston and Philly vs. 9% in NYC.

#21. Yields in the south and in New York were about 40% higher in 1900 than in 1890

We can combine the changes in sale and rental price indices to get at what happened yields, at least in proportionate terms (rather than in percentage point terms). Think of this as setting 1890 as the base year and then instead of applying just changes in one set of prices (sale or rental), apply both: suppose the yield starts at 100, and in the first year, rents rise by 5% but prices rise by 10%. In that case the yield falls by 5% (our index now reads 95).

Using this measure gives a clear picture of what happened yields in American housing in the 1890s. In NYC, the yield index went from 100 in 1890 to 139 in 1897; in the US South, it was at 130 in 1897. When the yield rises — in other words, when prices are a smaller multiple of annual rents — this signifies a fundamental change in financial/asset markets.

It could be that would-be buyers were just as optimistic about the future as they had been before, but lender behavior changed: what are termed non-price credit conditions. Or more likely it was some mix of changes in credit conditions and changes in (medium- to long-term) expectations: perceived future economic opportunities receded during the 1890s, in both NYC and the US South, pushing down prices relative to rents.

#22. In contrast to NYC, yields were unchanged in Philly and Chicago and fell in Boston

The findings in relation to yield for the largest city in the US and for the South — as noted above, intimate economic partners in the earlier 19th century — stand in contrast to the trend in other major cities. This is hinted at by the trends in the national figures: our estimate of gross yield rose slightly (from 8.3% to 9.0%) during the 1890s but not as dramatically as the change seen in NYC and the South.

Calculating the same yield ‘index’ for other cities highlights diverging fortunes. While yields rose sharply in New York during the 1890s, as people reassessed future prospects, yields were much closer to stable in Philly and Chicago, rising by about one tenth. In Boston, the other major city in the country at that point, yields actually fell slightly during the decade.

Share

#23. Housing prices in the West grew more twice as fast in the early 1900s as the national average

Housing prices in American cities, especially sale prices but also rental prices, fell during the 1890s. But they rebounded in the 1900s: both sales and rental prices rose by just under one quarter between 1900 and 1907. While these gains were broadly based, certain cities jump out — in particular in the (largely) new cities of the West.

Above, I mentioned that one of the challenges in building the dataset for the 1890s was about conventions in listings, rather than city size per se (for example New Orleans). But in the 1890s and as we enter the 1900s, an overlapping challenge emerges: new cities emerge, especially in the West, a few of which would grow into some of the US’s largest cities. Not counting Las Vegas, which only emerged after World War 2, the early HHP dataset covers six cities in the West: Los Angeles, Portland, Salt Lake City, San Diego, San Francisco and Seattle.

Listings are scarce in the earliest years for San Diego and Seattle. But the picture from the four Western cities is, nonetheless, clear: an index for the cities available, weighted by population, indicates that rents in the West increased by two-thirds 1898-1907, compared to just 20% nationally. And sales prices increased by 43% (vs 23% nationally).

This is true even if one subtracts out the impact of the 1906 San Francisco earthquake. The earthquake struck San Francisco in April and, by the time the fires had been put out, an estimated 80% of the city was destroyed and somewhere between half and three quarters of its population had been made homeless. Its impact in the rental market was, unsurprisingly, severe: markets rents are estimated to have been almost 50% higher the year after the earthquake than just before. But even setting this aside, as the graph above shows, housing prices in the West were rising faster than elsewhere during these years.

The strong gains in housing prices in these booming cities are telltale signs of supply struggling to catch up with demand. San Fransisco may have been an early mid-19th century leader but by the first decade of the 20th century, economic opportunities across the westernmost part of the US were being reflected in housing prices.

#24. Growth in housing prices in the early 1900s was uneven

Outside the West, a number of cities saw strong growth in housing prices in the decade or so to 1907 — but many cities did not. Detroit, for example, the emerging hub of the automobile industry, saw its rents surge by 70% between 1898 and 1907, behind San Francisco and Portland but ahead of LA and Salt Lake City. Memphis and Minneapolis continued their respective rebounds from the 1890s downturns mentioned earlier, with growths in rents similar to Detroit.

But in most of the older cities, the 1900s were a time of modest growth in rents. In Boston, for example, rents grew by just 12% between 1898 and 1907. New York, Philly and Louisville also saw less growth in rents than the typical city — while in DC, rents were actually 14% lower in 1907 than in 1898, driven in large part by a sharp contraction in rents 1898-1900.

New York saw sales prices grow stronger than rents in these years, meaning that yields fell back after rising in the 1890s. Many cities, though, with weaker rental growth also saw weaker growth in sale prices: in DC, prices rose by just 7%, in Boston 11%, while in Philly, they grew by just 2% between 1900 and 1908.

As the economic geography of the US changed, shifting westwards in particular, the so too went the housing market.

#25. The 1907 Panic led to a reversal of fortunes in rental market outcomes

Having risen for nine years in a row, rents peaked in 1907, a year marked by the next major Panic and recession in the US. Indeed, there is likely a connection between the San Francisco earthquake of 1906 and the Panic of 1907: huge volumes of capital were flowing west to help with reconstruction, around the same time that the Bank of England raised its interest rates. Without any Fed — indeed the Fed was set up in part as a response to the 1907 crisis — a scarcity of money ensued in New York, the financial hub. This was effectively an underlying weakness waiting for a trigger and the trigger duly came in late 1907, when the Knickerbocker Trust, the city’s third largest bank, went under after a failed financial gamble.

While largely a crisis of the financial sector, it spilled over into the real sector. Rents nationwide fell in 1908 and were largely static over the coming years. Indeed the average market rent in the US in 1914 was slightly below what it had been in 1907: seven years of plenty, seven years of famine, to borrow from the Book of Genesis.

But just as the run-up to 1907 was uneven, so too the fall-out of the 1907 Panic differed considerably across cities. Boston, which had ‘missed out’ on rising rents in the early 1900s, saw rents rise substantially after 1907. Rents in Boston grew by as much 1907-1914 as they did in the 1898-1907. In Philly, however, there was no such turn-around: rents fell 2% having risen by just 5% in the preceding decade.

At the other end of the spectrum is San Fransisco: as above, the 1906 earthquake destroyed the majority of the city’s housing stock. But while housing supply in the early 21st century is complicated, it was far simpler in the early 20th. A dramatic construction boom followed the earthquake and, by 1910, rents had come back down to their pre-earthquake levels. Of the 24 cities in the database in this period, San Francisco ranked 1st for increase in rents in the boom to 1907 and 24th (i.e. last) for the change in rents after.

But while San Francisco has a particular story, the reversal of fortunes — with rents falling post-1907 in cities where they had risen prior to the Panic — was not limited to SF. Baltimore also had a dramatic the turn-around: 46% growth in the build-up to 1907 was followed by a correction of 16%. So too with Cleveland, LA, Memphis, Minneapolis and New Orleans, while in Cincinnati, growth in rents simply came to a halt.

Next up

This is where we leave it for Post #3. In the Post #4, we’ll take a look at what happened in the sales market, and yields, in the run-up to World War 1 and then look at the immediate and medium-term consequences of the war — the biggest disruption to the American economy since the Civil War — on cities’ housing markets.

Our HHP series finish in 2006, because that’s around the time when listings, especially rental listings, start migrating from newspapers to online. (And we are working on adding the post-2006 period, as well as the 1870-1890 period.) But of course thinking about housing returns, from a timing perspective finishing in 2006, at the top of a housing cycle, is less than ideal. So below, I’ve extended to the end of the decade using the Davis-Lehnert-Martin series on rent-price ratio for US housing since 1960.

Why might we care about housing returns? If you need motivation, it is worth recalling that the return to housing (as a form of capital) is critical in understanding dynamics of wealth and wealth inequality much more broadly. In short, if you believe inequality matters, then housing matters. Arguably Piketty’s thesis about the rate of return to capital, relative to economic growth, hangs on what happens housing.

Rate of Return on Everything

One of the most prominent papers in this area over the last twenty years is a paper called ‘The Rate of Return on Everything’ (RoRoE), part of the stable of papers involving Òscar Jordà, Moritz Schularick and Alan Taylor. In addressing that key question of the relationship between capital and the rest of the economy, its innovation is that it includes returns on housing, rather than just focusing on more easily measured assets (in particular, sovereign debt and equities). Once housing is included, it sticks out a bit: across their dataset of 16 (mostly Western high-income) countries over the period 1870-2015, housing is a seemingly perfect mix of high returns (like equities) but low volatility (like debt).

But RoRoE also suggests that the US has one of the lowest (real, i.e. inflation-adjusted) returns to housing of the sixteen countries covered — and that this finding is driven by relatively poor capital gains, rather than by poor rental yields. More specifically, the US — with an estimated yield of 5.3% per year on average in RoRoE — ranks in the middle (8th) of the countries in the dataset. But only one country (Switzerland, 0.8%) has poorer capital gains in housing over the long run than the US (0.9%). This means, according to RoRoE, that the US ranks 14th of the 16 countries for share of housing returns that are due to capital gains (as opposed to rental yield) and 11th of 16 for overall housing returns.

For a country that likes to think of itself as a ‘Nation of Gamblers’ when it comes to housing — and where home ownership is a dominant political priority — this might come as a surprise. RoRoE uses adjusted versions of sale and rental price indices in the US that try to overcome the known limitations of those series. But what does this new dataset say? Let’s go looking…

Technical/Fun Stuff

But first, some technical notes! Before we get started, we need to make a few choices (five main sets of choices, to be precise) to convert from a concept of returns to housing over a long time period to numbers. If you are not a fan of technical notes, you can skip the bullet list and go to the paragraph below, starting ‘So, in sum...’.

1. Components: Firstly, as mentioned above, there are two main components to returns to any asset, including housing. The first component to returns is the income it generates, often called the dividend or, in the case of housing, the ‘rental return’ (RR; see Note 1!). The second component is the capital gain (CG), in other words how much the price (or value) of the asset itself has changed. The total return (TR) to housing in any given period is the sum of the two components, TR = RR + CG.

2. Gross vs net: Secondly, there is a distinction between gross and net returns. One of the most salient drivers of the gap between the two is taxation. But the challenge is that the taxation of housing returns (either component) varies over time, by location, by housing type and indeed by price point (think: mortgage interest relief). It may also vary by owner type (think: rental income accruing to a REIT versus a ‘naive’ mom-and-pop landlord). Because of the sheer combined strength of assumptions we would need to make along all those dimensions, we don’t attempt to adjust our gross returns — instead, we simply present gross returns (at least for now; RoRoE does attempt to make this adjustment). This also means that our returns do not account for maintenance or depreciation. This is not unusual in long-run studies of housing, by the way, but that does not mean we should forget that we are doing that.

3. Nominal vs real: The third main note to make is that we can think of both nominal (‘sticker price’) and real (adjusting for inflation). As per the literature, such as RoRoE, we adjust for inflation after calculating total returns, in particular thinking about real capital gains, rather than nominal. Nominal can matter, though… see Fact #16!

4. Geometric vs arithmetic: A fourth technicality to consider is whether we are interested in average returns (assuming some long horizon for investing) or the experience of the average year. Suppose there were just six years in our dataset and in five of those years prices rose by 10% but in the other year they fell by 50%. With five +10%s and one -50%, the simple average of those is 0% (the five 10s and the one 50 cancel each other out). But if we had held the home as an asset for all six years, it would be worth almost 20% less at the end than the start: this is the power of compounding (here in reverse, with a disproportionate fall). In the paper, we focus mostly on the average year, again matching RoRoE. But below, I'll be a little more ‘financial’ and mostly focus on (geometric) returns, as would be experienced by an investor.

5. Periodicity: The last thing to think about/choose is the periodicity with which we calculate housing returns. In general, people think about returns on an annual basis. But when thinking about returns over the medium- to long-term, which I take below to mean returns over ten years or more, it is worth rethinking those calculations, in particular for yields. To see why this is the case, suppose our desired periodicity was monthly. In that case, to calculate monthly rental yield, we would need to recalculate the denominator each month: how the price of the home has changed since a month ago would need to be factored in. Thinking about ten-year returns is basically just switching it the other way around: if, at any point, someone bought a home and held it for ten years, the rental yield that they enjoy is off the base from when they bought it. So, when thinking about ten-year returns, rather than annually, we basically need to recalculate the return to reflect that. (As in: someone who holds a property for ten years, in a period of rising prices and rents, doesn’t face the higher prices in any subsequent year but does get to enjoy the higher rents.)

So, in sum, the focus for this post will be on both components of housing returns, in gross terms, but adjusting for inflation (especially in capital gains) where noted. I will mostly present geometric returns (rather than arithmetic averages of returns) and the typical ‘holding period’ under consideration will be ten years, although (in particular just below), I’ll also present year-on-year returns.

Subscribe now

And with those prelims out of the way, off we go! Facts #1-#8 are back in the first post in this series, so we start today at #9…

#9. US housing returns average 8.6% per year, after inflation

Over the period 1890-2010, the typical year saw housing values rise by 3.8% — unsurprisingly lower than the 4.2% if we stop at 2006. Compounding affects this and with substantial nominal gains earlier, the AGR — that is, the geometric average that would be enjoyed by someone who held property for the full span — is a lower, at 3.6%.

However, inflation matters when considering the returns to housing. The typical year in our period saw inflation of 2.8%, although it was uneven. (We’ll return to patterns over time later.) Subtracting this from the measure of capital returns gives an estimate of the average long-run return to housing in the US of 8.6%.

Of course, on top of this, anyone who held housing would have enjoyed rental income as well. The HHP series estimates an average rental yield across the period of 7.6% (gross). Putting the two together gives a nominal return to housing of 11.4% for the typical year.

The first figure in this post, above, gives the return by year and by component, in real terms. In only nine of the 120 years was the return to housing negative — and four of those came 2007-2010. In contrast, in one third of the years covered (45 to be precise), the return to housing was over 10%. As I outline in more detail below however, this prevalence of strong returns, even after inflation, was due to the strength of rental income, rather than the strength of capital gains.

#10. The 1940s, the 1920s and the 1970s saw the strongest ten-year returns to housing

The graph above shows how things changed year-by-year. But of course real estate is a long-lived asset and, related, the holding period is usually going to be longer than a year. The typical dwelling in the US changes homes something like once every ten years so we can tweak the calculations to see which ten-year period offered the greatest returns.

To recap on this subtle but important difference, let’s think about a period of high inflation, when both sale and rental prices are rising rapidly (say, the mid-1960s to mid-1970s). Our investor buys at the start of this period and holds for ten years. This means that they do not have to pay the increases seen in sale prices; indeed, it’s the opposite as they will enjoy those substantial capital gains when they sell. But, because they are renting at market rates, they do get to enjoy the increase in rents.

If we are thinking about 1965-1975, the investors gets to enjoy 1975 rents off a 1965 base. For those curious, the value of a home rose 95% between 1965 and 1975 while market rents rose 73%. While the rental return as calculated above would have fallen slightly in this decade (as prices rose by more than rents), for our investor-in-their-tenth year, they are enjoying a yield that is 1.73 times what it was in their first year.

The second figure in this post, below, shows inflation-adjusted returns to US housing, by decade from the 1890s to the 2000s, using a ten-year holding period (rather than year-by-year calculations). The numbers are expressed in annual growth rates and, as above, are broken into the two parts (rental returns, capital gain) with total returns also shown (this time in black diamonds).

Share

In each of the twelve decades, the return to housing was positive: that aligns with the black line in the first figure above being positive almost all the time. In all periods, similar to the RoRoE finding mentioned above, it is rental income, not capital gains, that dominates.

But what is interesting about this figure is that it shows the heydays of housing returns to be somewhat different to the popular perception that housing became a uniquely profitable investment after the 1960s. Indeed, while three of the five lowest-returns decades of the twelve are in the first half-century (the 1890s and the 1910s saw prices fall in real terms, as did of course the 1930s), the other two are the last two (the 2000s, 11th of 12 decades) and the 1990s (9th) — with the 1980s next lowest.

The three decades in which housing performed most strongly are, in order, the 1940s, the 1920s and the 1970s, with the 1900s not far behind. Now some of this is timing — the decade to 2006, if one were to stop at the top of the bubble, looks quite different to the decade to 2010 for obvious reasons. But let’s return to that later.

#11. Rental yield dwarfed capital gains in housing returns

The overall figures show just how much more important rental yield was, relative to capital gains. Over the full 120-year span, rental income made up over 90% of aggregate housing returns.

Strictly speaking this figure is not directly comparable with the RoRoE figures for other countries, because those figures make assumptions about tax treatment and maintenance to come up with an estimate of net returns. As above, though, the ratio of gross to net returns varies in many dimensions — a homeowner ‘renting out’ their property to themselves doesn’t pay any income but a landlord renting out to someone else may, depending on their tax strategy. Thus, for simplicity, we focus here on gross yields.

But even with a crude assumption of net-to-gross income equaling 80%, this would still mean that, over the full period 1890-2010, rental yield made up roughly three-quarters of all housing returns.

#12. Rental yields rose to the 1920s and fell from the 1940s

From a careful look at both the graphs above, it should be clear that rental yield — the annual income produced by a home, even if only imputed in the case of owner-occupiers — is at the heart of housing’s high returns. In this way, the experience of the US is consistent with the experience in other countries, even if the share of return due to yields rather than capital gains is perhaps a bit higher.

While that was certainly the case over the 20th century as a whole, a closer look at rental yields indicates that its contribution to housing returns is not stable over time. Between the 1890s and the 1920s, housing yields grew. In the interwar years, it fell in the 1920s but rose again in the 1930s, just as one might expect in a classic bubble/crash episode.

But from WW2 in the 1940s through to the Global Financial Crisis of the 2000s, the yield on housing steadily fell. This is true regardless of holding period, as the third figure in this post shows. Across one-year (blue; the baseline), five-year (orange) and ten-year (grey) holding periods, yields peaked in the 1920s, rebounded in the 1940s and fell — albeit with a pause in the 1960s — until the late 2000s.

In the 1890s, estimated gross yields were about 8% but this rose in the post-WW1 period. The annual yield peaking in 1920 at 14% gross — although this was predicated on market rents, at a time when cities were still coming out of various post-war rent control systems. By 1930, however, with rents falling and prices rising, yields were just 7% — only to yo-yo back up above 10% by the mid-1940s, as another war/postwar spike affected rental market outcomes.

From a peak of 11.1% in 1946, yields fell decade on decade and in 2006 were less than one third of what they had been sixty years before, at 3.4%. Is it any wonder that the 1990s and 2000s had, relative to other decades, poor returns to housing an asset when yields were so low?

#13. Capital gains are ten times as volatile as rental yield

Let’s look next at real gains in housing prices in the US since 1890. The first thing to note is that capital gains are volatile. One way of measuring how spread out a dataset is to look at the range: what are the minimum and maximum values? Another is known as the ‘coefficient of variation’ (CV), which is the standard deviation expressed relative to the mean. (Crude rule of thumb: if the standard deviation is bigger than the mean, making the CV greater than 1, it’s a spread out dataset.)

To set the scene, rental yields — the other main component of gains — had a range over that period of 3.3% (in 2007) to 14.3% (in 1920). With a mean of 7.6% and a standard deviation of 2%, its CV was 26%.

The equivalent figures for capital gains are a different order of magnitude entirely. While the range for rental yield was nine percentage points, the range for real capital gains was 44 percentage points: from a 14% fall in 1932 to a 17% rise in 1946.

With a mean of 1% but a standard deviation of 5.2%, the CV for capital gains over 120 years is 523%, about twenty times that for rental yield.

#14. The 2000s housing bubble was different: it saw the strongest real gains on record

One of the key messages in Shiller’s book ‘Irrational Exuberance’ was — at least in its second (and subsequent) editions — that the housing price growth of the late 1990s and early 2000s was unique, in historical perspective. As the analysis in the paper and above hopefully shows, there was nothing new in strong returns to housing as an asset and, taking the 2000s as a whole, its housing returns do not stand out.

But the 1990s and 2000s were different in one key way. The decade to 2006 saw the strongest real capital gains of any ten-year period in the dataset. Looking back in 2006 over the previous ten, and stripping out inflation, the average property owner would have calculated their gain to be 6% per year, every year, for a decade. As the fourth figure in this post shows (below), that is the highest ten-year average in the entire series. This may seem at odds with Fact #10 above, but that is because there we were taking each decade on its own terms (including the fall in prices 2006-2010) but here we are calculating returns for each ten-year period.

Subscribe now

The ten-year period with the second largest gains was the decade to 1928, when the value of homes rose by 5.6% on average, above and beyond inflation, each year for a decade.

In both cases, though, strong capital gains seem to have stretched the relationship between prices and rents beyond breaking point. Yields halved 1920-1929, from 12% to just over 6%. If they were ‘too high’ in 1920, reflecting a housing ‘famine’ (as one commentator termed it at the time), yields were ‘too low’ a decade later and the subsequent fall in prices, by 27% in real terms, pushed yields close to 9%.

#15. The ‘Great Recession’ just pips the ‘Great Depression’ as the worst housing crash on record

Similarly, in the 2000s, yields fell ‘too low’, bottoming out at just 3.3% in 2007, before correcting to around 4.5% in the early 2010s (per the DLM dataset mentioned above). As Barry Eichengreen noted in his ‘Hall of Mirrors’, there are a number of parallels between the 1920s/1930s boom and bust and its 2000s equivalent.

The two episodes mark the biggest and second biggest upswings in prices (after inflation) since 1890. But, having peaked in 1928 in one case and in 2006, prices fell sharply — even with wider deflation, especially in the 1930s episode. The falls were quite similar, after inflation: between 1928 and 1934, the average price of a home in the US fell by 27% after inflation. Between 2006 and 2010, it fell by 28%.

#16. Nominal returns, which matter for mortgages, peaked in the 1980s

Everything presented so far has been dutifully adjusting for inflation. In general, when considering returns to any asset, it’s important to take account of wider inflation, which eats away at returns. But nominal capital gain can be important, too, if there is leverage or debt involved.

And over the course of the 20th century, housing went from being a typically rented service to one that was owned, with leverage, by the typical household. With mortgage debt fixed in nominal terms, this means that high inflation may reduce capital gains on paper but if consumer prices trebled (as they did 1964-1982), housing prices (more than) trebled in the same period, but your mortgage stayed the same, then you were significantly better off.

The final chart of this post adds back in wider inflation and sticks with a ten-year perspective to give a sense of what an actual property owner might have experienced. Taking 1950 as an example, in the ten years to 1950, there had been an average inflation-adjusted gain in housing prices of 3.3% per year. But prices had also gone up by 5.7% per year, on average, in the same decade. This meant that the ‘sticker price’ in dollars for the home had increased by an average of just over 9% per year for a decade.

If you had taken out a modern-style mortgage in 1940, it is not just the 3.3% that matters. The 9% matters because the principal of your mortgage does not go up with inflation (and thus neither does its repayment, if — as is standard in the US — the interest rate is fixed).

Viewed in these terms, every single ten-year period in the postwar years would have led to the conclusion that housing was a ‘good bet’. This was even true in the late 1990s, when adjusting for inflation, prices nationally were lower than a decade before, because inflation itself was substantial. In the decade to 1997, real prices fell by an average of 0.9% per year but, with inflation of 3.5% per year, the nominal returns have the opposite sign, with an average gain of 2.6% per year. Low, certainly, compared to other periods, but still positive and relevant for those with mortgage debt.

This long housing boom, in nominal terms — from the 1940s to the 2000s — likely reshaped the economics and political economy of the housing system in the US. Land-use regulations are perhaps the single biggest driver of housing unaffordability in the US in the 21st century.

And as Bill Fischel described in his excellent 2015 book, Zoning Rules!, we need to think not just about the supply of these regulations but also about the demand for them. The 1970s saw record inflation, which spilled over into housing. For all its economic costs, it brought benefits to homeowners in the form of drastically reduced mortgage debt burdens. The 1970s also saw the birth of the current generation of land-use regulations, even if there were deep roots that went back almost a century.

There is a plausible link between returns and zoning. But each city had its own experience. In the next post, the third in the series, we will start to explore city-level housing market dynamics, using the database of thirty cities covered.

Subscribe now

Until now. As of June 2024 (and updated substantively in August 2025), the new Historical Housing Prices Project, hosted by the Federal Reserve Bank of Philadelphia, is live. This is a project I’ve been working on for almost a decade, with my collaborators Allison Shertzer and Rowena Gray — and with financial and logistical support from the Lincoln Institute for Land Policy, the National Science Foundation and the National Bureau of Economic Research as well as Trinity College Dublin, the University of Pittsburgh and UC Merced.

It helps to keep one map in mind: the paper divides 116 years into three acts — a flat, cyclical era to 1940, a stable mid-century (1940–1970), and a fast, divergent era after 1970, when sale prices and rents permanently parted ways. These posts walk through them in order.

What? And why?

The goal starting the project was, at least in theory, relatively simple. Surely it should be possible to generate reasonably accurate indices of housing prices, both sale and rental, for major US cities for the 20th century. The fact that these didn’t exist struck me as odd, given how central housing is to modern economic life. (This is true even if you don’t subscribe to the ‘housing theory of everything’.)

OK, you say, but — as an investor might say — why now? And why you?

Over the last ten years, millions of pages of historical newspapers have been digitized, principally with genealogical goals in mind. But newspapers contain more than just records of people’s grandparents. Real estate listings were the backbone of newspaper’s golden age in the U.S. Indeed, it’s somewhat ironic that, while conventional wisdom blames Facebook and Google for eating the revenues of the major newspapers, really it was real estate listings moving online that did the damage. (If that sounds like a stretch, consider the fact that the real estate section of the LA Times in the 1990s had almost 100 pages!)

So the 2010s brought about the opportunity to excavate reams of data that lay hidden in the microfilms and microfiches of libraries around the US.

Who?! And how?

But that still doesn’t explain how I got involved. After all, I’m an economist based in Ireland. As it happens, my areas of interest are housing markets and economic history. I’ve been working in both spaces for two decades, including extensive work converting real estate listings into housing price indices, through my work with the Daft.ie Report in Ireland and in work for the IMF with a range of countries around the world. The logic carries over from online listings to newspapers listings: listings have both a measure of price but also measures of size, type and location, meaning it is possible to come with reliable measures of property price trends over time.

So, with Allison, Rowena, and a huge team — there is a literally an appendix of acknowledgements and thank yous — we gathered listings for each of thirty cities, as far back as they would go, to 1890, and up to 2006, at which point online listings start to take over. There are lots more technical details in the NBER Working Paper, joint with our co-author David Agorastos. But in brief, we built a dataset of 2.7 million listings for our thirty cities, which were chosen to cover as many regions and economic trajectories as possible. Controlling for housing unit size, type, and location in city from the ads, we use hedonic regressions to construct indices at the city level, at annual frequency. A “rolling windows” method helps us to minimize the challenge of unobserved changes over time, including in the quality of housing and in the price of locations.

A note on coverage: most cities enter in 1890, but some appear only once their secondary market was deep enough to advertise widely — Baltimore from 1908, Tampa 1905, Miami/Seattle/Phoenix around 1910, Las Vegas only after WWII. So the earliest decades rest on a dozen-plus cities; the full 30 are in place by the late 1910s.

What’s new?

Over the next few weeks, my plan is to delve into this deep database and highlight dozens of new insights. Today’s post will go through the first eight headlines, all of which are national in nature, across both sales and rental segments. The next post will look at what we know from combining both segments: the return on housing as an asset. After that, future posts will go through the various main periods and cycles in the US housing market since 1890 and pick out key new insights.

Subscribe now

#1. The real price of a home in the US was almost four times higher in 2006 than in 1890

Until this HHP dataset, the best information we had on how the sale price of housing evolved in the US from 1890 was the Shiller index, first published in Bob Shiller’s book Irrational Exuberance in 2000. As we explain in our paper (and indeed as Shiller himself explained in his book), there are numerous reasons to believe there may be certain limits to this pioneering index, based on its underlying sources.

And indeed one of main findings is that there was more growth than previously thought in housing prices over the course of the twentieth century. This is seen either in ‘real’ terms (stripping out inflation) or in nominal terms (the actual number of dollars in each year). Start-to-finish, real prices were 3.9 times higher in 2006 than in 1890 (vs. 2.25 times in Shiller’s index) — in nominal terms, 89 times rather than 51 times.

In addition to revising the top-line finding, there is also important variation in the trends over time, in particular for the periods up to the mid-1980s, when there is uncertainty about the underlying data in the original series.

For the earliest period (1890-1914), indeed, the Shiller index rises more than the HHP index does: in nominal terms, (27% vs 18%). And, as noted above, the 1987-2006 change is very similar: 192% vs 185%. (This latter similarity lends credence to the overall exercise as economists and statisticians largely trust the sales price indices after 1987.)

This means that the differences between the two series stem from the considerably shorter 73 years between 1914 and 1987, rather than the full span. For this shorter period, 1914-1987, Shiller’s index increases in real terms by just 18%, while — according to our new index — the sales price of housing rose by 126%. In AGR terms, this means that real housing prices rose more than four times than previously thought for the period 1914-1987: by 1.1% per year on average, not 0.2%. This is a substantial revision to what we thought we knew about housing prices.

Share

#2. Rents have risen, not fallen, since 1914

On real prices, the previous estimates were the right ‘sign’ but the wrong size. On real rents, though, our work shows an even bigger discrepancy. There is very little annual market rental data for any city in the United States until after 2000, so everyone has until now used the BLS Rent of Primary Residence (RoPR) series, which is based contract rents and is used to compute the CPI.

But, as we explain in the paper, there are huge question marks about this series. As early as the late 1940s, the BLS’s own statisticians were concerned that the RoPR series was understating rental inflation but this wasn’t fully fixed until the early 1990s, when the methodology was changed. We are not the first, by any means, to state this. Crone, Nakamura and Voith — among others — have done sterling work trying to adjust rental series to get a more realistic trend.

However, what we can do that others could not is bring new data. These are, of course, listed rents, not contract rents, but we believe that trends in listed rents are, when handled well, highly correlated with trends in underlying market rents. And, just as with sale prices, we find far more inflation than previously thought. We find is that (nominal) rents rose by a factor of 28, not a factor of 11, between 1914 and 2006.

The RoPR series suggests that, adjusting for inflation, rents fell by 36% between 1914 and 1948 and continued to fall, albeit more slowly, thereafter — ending in 2006 at about half their 1914 level. We find, instead, that rents were largely the same in 1945 as they had been in 1914 and that they rose, around a number of major cycles, thereafter finishing in 2006 at 60% above their 1890 level, rather than 50% below.

In AGR terms, real rents rose by an average of 0.4% per year over those 12 decades, rather than falling each year by 0.6%, on average. This is a substantial change in our understanding of the long-run path of rents — with implications for the wider cost of living, discussed more below.

#3. The Twenties were indeed Roaring

Shiller’s index of sale prices is based on five sources spliced together. The earliest source, covering the long period 1890-1934, is an index based on a survey of homeowners conducted in 1934, by housing hall-of-famer Leo Grebler together with his coauthors Blank and Louis Winnick (also a prominent housing researcher in his day). Their survey asked homeowners what they recalled paying for their home and when they bought it, as well as what they believed their home was worth at that time.

You don’t need to be a behavioral economist to think about potential sources of bias in this set-up: from bunching in year of purchase to rounding in recalled purchase price, but also challenges in understanding the true value of a home at the height of the Great Depression as well as no correction for home improvements since purchase.

But these limitations are hard to sign — which is econo-speak for, we don’t know whether it biases things up or down. It turns out the major flaw in the survey data was not about what statisticians call the ‘first moment’ (i.e the average) but instead about the ‘second moment’, the variation in the series over time. Relative to what the HHP project reveals, the survey was simply too stable over time, resulting in a forgotten boom (and bust; more below).

That ‘Roaring 20s’ boom can be seen in both sale and rental prices, although the timing differs. As can be seen below, market rents more than doubled between 1916 and 1920 and, while they fell back over the following six years, they remained at roughly twice their 1914 level. You can see this again and again, in individual listings and at the city level (using hedonic regressions, adjusting for the mix of homes listed).

While rents rise in the RoPR series, the increase is slower (itself not a huge concern given the different path of contract and market rents over time) and only about two thirds as much, compared 1926 with a decade earlier.

The survey data seems to completely miss the housing boom that followed World War 1. Part of this was immediately postwar, and so likely related to returning troops and the ‘demilitarization’ of the economy: while Shiller index (and its underlying Grebler et al component) only rises by 20% 1918-1920, the HHP series increases by over 50% in the same two years.

But while market rents had peaked in 1920, sale prices were only starting their rise. They continued to rise, growing a further 25% 1920-1926 to leave them over twice their 1914 level at their peak. The Shiller index, however, shows no growth in the same six years, meaning almost all of a doubling of housing prices has been missed until now.

(For now, I’ll just note these trends. The point of these posts is to highlight new facts, with some musings on potential drivers, rather than to outline clear cause-and-effect. However, what’s worth noting across this point and the next is the classic Kindleberger bubble sequence of fundamentals → asset prices → wider economy. Rents peaked in 1920-23, sale prices in 1926 and the wider economy in 1929.)

#4. The Great Depression happened (in housing)

The Great Depression, to paraphrase Ben Bernanke, gave birth to macroeconomics. Until then, the dominant mode of thought about the economic system was that it was largely self-correcting. Since then, the dominant mode of thought has been that policy is a necessary ingredient in a healthy economy.

As Barry Eichengreen outlines eloquently in his book ‘Hall of Mirrors’, housing was at the heart of the run-up in the Roaring Twenties and thus at the heart of the crash that was the Great Depression. I’ll go into some of the city-specific stories around these decades in later posts but what’s reassuring — especially to people like Price Fishback, who has been making this point for some time now — is that the run-up before and crash during the Great Depression is there in the HHP series. It hadn’t been so obvious before.

Shiller’s index does show a fall in prices, of just over 25%, but that fall only matches wider deflation in the US economy, meaning real prices in 1934 were just 1.6% lower than in at their peak in 1928. This is confusing because the RoPR series — at this point based on surveys of landlords and lettings agents — shows rents falling by about one third, i.e. by 10% more than the fall in wider prices.

I return to this later but it seems hard to believe that the yield (the annual rent as a share of the value of a home) went down, not up, during the greatest crash in US history. A bubble should be marked by yields going down — people are happy to bid more and more to buy a home, relatively to the underlying income it gives — and then the crash should be marked by yields recovering.

Share

And that is what the HHP series shows. The fall in rents in HHP matches the fall in RoPR: nominal rents fell by a third. But sale prices fell by more than rents, not by less. In nominal terms, sale prices were over 40% lower in 1934 than they had been in 1928, at their peak. Adjusting for falling prices, there were over one quarter lower.

Yields went up in the Great Depression, not down; more below.

#5. The missing rent spike

The value of housing took a long time, and a World War, to recover after the Great Depression. Prices peaked in the mid-1920s at just over twice their pre-WW1 level but a decade later were just 10% above that 1914 level.

Similar to during and particular after World War 1, the Second World War saw inflation once again in all prices, including housing. Between 1941 and 1948, consumer prices rose by at least 7% each year — and by 25% in one year alone, 1946. Overall, in these seven years, housing prices more than doubled (116%). This matches the Shiller index, which between the mid-1930s and the mid-1950s is based on median prices from (guess what!) newspaper listings in five cities.

In the rental segment, however, the existing series — the RoPR series — implies that there was almost no growth in rents in this period. In the period 1941-1948, nominal rents in the BLS series rose by just 14%. But with the wider price level increasing by two thirds in the same period, this would mean that in real terms rents fell by nearly one third.

If true, it would mark the start of the long period of decline in real rents over four decades, until the 1980s. But the HHP series suggests that the rental sector saw an increase in prices almost as large as the sale segment did, with nominal rents more than doubling (up 121%; real rents up 35%). We discuss in the paper that contract rents and market rents are, of course, not the same thing, but they should be going in the same direction.

Further, while there is some debate about this, researchers such as Adam Ozimek have argued market rents are also the relevant object when thinking about the cost of housing for owner-occupiers. (More below.)

If it were a one-off, perhaps you think of this as a blip. But, as per #7 below and the BLS economists themselves at the time, this was the start of something more systematic. I think it is more accurate to say that real rents rose significantly — almost by as much as sale prices — during and immediately after WW2 than to say that they collapsed.

It’s also worth noting that both World Wars had similar impacts on housing markets. This is something (narrative) historians have talked about a lot but economists, economic historians and others simply haven’t had the data to see, until now.

#6. Between 1950 and 1985, home prices did not flatline — they rose by two thirds

So far we have seen, in both sale and rental segments, the HHP series imply significant revisions to the path of housing prices over time, compared to what we thought we knew before. I say ‘what we knew before’ but, in line with a few of the references and links above, we are far from the first to worry that existing series are not fully accurate. (In fact, you can take our paper to be a response to these existing concerns, rather than raising them from scratch.)

But it is in the generation after World War 2 that the two most significant challenges to existing series occur. On the sales side, as Greenlees wrote in the 1980s — and as Morris Davis has written about extensively since, bringing Census data to bear — there are concerns about the methodology and data used to generate the sale price component of the CPI. In addition, for the period 1975-1987, the OFHEO measure of prices includes appraisals, which may bias prices down. (For more, see the paper.)

What muddies the waters here is substantial inflation between the mid-1960s and the mid-1980s, so we’ll focus on real measures. According to existing measure of sale prices, real sale prices were about 4% higher in 1985 than in 1950. But the HHP measure indicates that prices in 1985 were two thirds higher than in 1950, rather than effectively the same.

#7. Between 1965 and 1980, market rents did not fall 24%, they rose 15%

According to the RoPR, the falling trend in rents, over time, continued after the 1940s all the way into the early 1980s, with rents in 1981 apparently just 46% of their 1914 level.

However, the HHP dataset indicates that no such fall happened in market rents. Instead, rents had a modest upward trend, with cycles for much of the three decades after World War 2. There were five broad rental market cycles in the US between 1945 and 2006, with rents spiking in 1948 (as above) but peaking again in 1969, 1980, 1986 and 2001.

The period 1965-1980 is perhaps a critical one when thinking about the wider implications of these revisions to our understanding of housing prices over the long 20th century.

The BLS RoPR measure suggests that rents fell by one quarter in these fifteen years. But market rents rose by 15% in the same period, according to our data and mix-adjusted methods. There are two bits of important context here. The first is that inflation (in broader consumer prices) was high throughout this period. The second is that rents often “reset” between tenancies. The method of capturing rents was based on surveying households, not dwellings, meaning a ‘non-response bias’ just at the time when rents reset to the market would systematically bias down the measure of rents over time.

The fact that our HHP rent index differs from the RoPR measure just when inflation is at its highest is consistent with those, such as Bob Gordon and Crone-Nakamura-Voith (as mentioned above), who have long argued that rents have not been falling, decade-on-decade, for a century.

The final two charts show the new and old indices, in real terms, for both home sale prices and rents, from 1890 to 2006 (here indexed to 1890 = 1). The gap between previous estimates and new estimates in both segments is obvious — as is the cyclicality of the new rent series around the upward trend.

#8. All this means that inflation was likely greater than previously thought

So far, we have talked about ‘real’ as well as nominal housing price indices, where we use the Consumer Price Index (CPI) to scale things so that they are comparable across long time spans. However, if the measure of housing prices is wrong, then so to is the measure of CPI.

This is something Allison, Rowena and I are coming back to in its own paper, which will be presented at a special NBER conference in Measurement in Housing in March 2026. But in short, if (nominal) housing prices increased by more than we thought until now, then so too did the measure of all consumer prices — because housing/shelter is such a key component of what households spend their income on.

Subscribe now

What’s next?

That’s hopefully enough for one sitting! In my next post, I’ll look at what we now know about housing as an asset and the returns to housing (both capital gains and rental yield). I hope that you enjoyed this and will come back for Part 2!