Methodology
The source
The Jorda-Schularick-Taylor Macrohistory Database, Release 6: 18 advanced economies, 1870 to 2020, 2,718 country-years. FinObservatory reads the released file and recomputes. No figure on these pages is transcribed from the published paper, because a number typed from a paper is a number nobody can check against the data, and this repository does not publish those.
Returns are nominal in the source, and must be deflated
The return columns are nominal total returns, expressed as decimal rates. They are deflated here by the database's own consumer price index:
real = (1 + nominal) / (CPI_t / CPI_t-1) - 1This is not a stylistic preference. German equities returned 2,639,074,816 in 1923, in nominal terms, because the currency was collapsing. Average the nominal series and the headline becomes a hundred-million-percent return. Deflated, that same year is a real gain of about 150 percent, which is a statement about an actual event rather than an artefact of the unit of account.
The average is geometric
Returns compound, so the mean reported everywhere on these pages is the compound annual (geometric) mean of the real gross return, not the arithmetic mean. The arithmetic mean overstates what an investor actually earned, and it overstates it by more when the series is more volatile, which would systematically flatter exactly the assets these pages are trying to compare honestly.
The war years
1914 to 1918 and 1939 to 1945 are in the full sample. They were real years and real returns, and dropping them silently would be a choice made on the reader's behalf without telling them. But an average that quietly contains them is a different statistic from one that does not, so both are computed and both are shown. Equities return 4.43% over the full sample and 5.30% with the wars removed; the gap is the war.
What the housing series cannot tell you
This is the most important limitation on these pages, and the one most likely to mislead. Housing shows a higher real return than equities at less than half the volatility. Before drawing any conclusion from that:
- The volatility is understated, by an unknown amount. House price indices are smoothed by construction. Houses trade rarely, and valuations lean on appraisals and on comparable sales that are already stale. The index moves less than the asset does. This data cannot measure by how much.
- Nobody holds the index. It is a diversified national portfolio of all housing. A real household owns one house, on one street, exposed to that street. That idiosyncratic risk is invisible here.
- Costs are not netted out. Transaction costs, maintenance, property taxes, vacancy, and the illiquidity of a house are not in these returns. Equities have none of those to the same degree.
- Leverage is not modelled. Housing is typically bought with a mortgage, which changes both the return and the risk of the position, in opposite directions.
The housing column is the return on the national housing stock. It is not the return you would have earned, and these pages do not claim it is.
Coverage
16 of the 18 countries in JST have return data. Canada and Ireland are in the panel but have no return observations of any asset, so they have no page. 39 observations across the panel carry JST's own interpolation flag: they are estimates, and they are included rather than dropped.
| Country | Equities | Housing | Bonds | Bills | Interpolated | Crises |
|---|---|---|---|---|---|---|
| Australia | 151 | 120 | 121 | 148 | 0 | 2 |
| Belgium | 151 | 120 | 145 | 147 | 16 | 8 |
| Germany | 151 | 116 | 146 | 145 | 0 | 5 |
| France | 151 | 150 | 151 | 144 | 0 | 4 |
| Italy | 151 | 86 | 151 | 131 | 0 | 9 |
| United Kingdom | 150 | 118 | 151 | 151 | 0 | 4 |
| Portugal | 150 | 73 | 150 | 141 | 3 | 5 |
| Sweden | 150 | 138 | 150 | 151 | 1 | 6 |
| United States | 149 | 130 | 150 | 151 | 0 | 6 |
| Denmark | 148 | 145 | 150 | 146 | 0 | 6 |
| Norway | 140 | 150 | 151 | 151 | 0 | 4 |
| Japan | 133 | 75 | 140 | 145 | 15 | 7 |
| Finland | 125 | 99 | 151 | 151 | 0 | 5 |
| Switzerland | 121 | 119 | 120 | 151 | 0 | 5 |
| Spain | 121 | 120 | 117 | 147 | 4 | 8 |
| Netherlands | 121 | 150 | 151 | 151 | 0 | 2 |
| Canadano return data, no page | 0 | 0 | 0 | 0 | 0 | 1 |
| Irelandno return data, no page | 0 | 0 | 0 | 0 | 0 | 1 |
Crisis years, and why they come from JST
The crisis bars and the crisis-year statistics use JST's own crisisJST flag, not the FinObservatory crisis atlas that powers /crises. The two chronologies date some episodes differently. A crisis-window return computed against a different chronology than the table it is joined to would not reconcile with the rest of this site, and a number that cannot be reconciled is a number waiting to be wrong.
The crisis statistics are descriptive. They report what these assets did in years already known to have been crises. They are not evidence that any asset predicts a crisis, and nothing here should be read as a forecast.
Why the series stops in 2020
Because the source stops in 2020. FinObservatory holds a BIS residential property price index that runs later, and it is tempting to splice it on. It is a price index with no rent leg, and a total return is price plus income. Extending a total-return series with a price-only series would manufacture returns that nobody measured, so these pages end where the data ends.