How hot and cold are measured

1. Position: each indicator is compared only with its own past

For every indicator, this site asks the same question: is it now higher or lower than its own normal level in the past? There are three methods, depending on the kind of indicator:

• Prices that grow over time (share prices, exchange rates, commodities, house prices): this month’s value is divided by its own average over the last 10 years, and the ratio is then ranked within its own history. A ratio of 1.2 means 20% above the 10-year average.
• Interest rates and growth rates (policy rates, government bond yields, inflation, and year-on-year growth in GDP, employment and wages): the average over the last 10 years is subtracted from this month’s value (the gap in percentage points), and the gap is then ranked.
• Indices that are centered on normal by design (confidence, manufacturing surveys, financial stress, supply chain pressure, the yield curve, valuation, the business cycle signal and others): the value itself is ranked directly.

The rank is given as a percentile, like a percentile rank on an exam: 72 means higher than 72% of its own past months; 50 is its own normal level. Quarterly data (GDP, house prices, the bank lending survey) use the same method, counted in quarters.

2. Five levels from cold to hot, and “at a historical extreme”

The five levels are divided at 20, 40, 60 and 80: below 20 cold, 20 to under 40 cool, 40 to under 60 neutral, 60 to under 80 warm, 80 and above hot. At 90 or above, or 10 or below, an indicator is “at a historical extreme”: the outliers this site sets out to catch. For most indicators such a position is uncommon, and not necessarily bad; indicators with a long upward trend (such as US valuations) have often sat at the high end since the 1990s.

Hot and cold always mean “above its own normal = hot, below = cold”. They describe position only, not good or bad: unemployment above normal is also called hot (usually a sign that the economy is weakening), and a yield curve below normal is called cold.

3. The normal level uses no later numbers

The position for every past month is measured against a normal level built only from data up to that month: a position in 1990 is compared only with data from before 1990. This keeps the yardstick free of hindsight, but the values themselves are today’s published versions (GDP, jobs, prices and leading indicators are revised later) and publication delays are not taken into account, so they are not exactly what people saw at the time. An indicator needs at least 10 years of ratios (120 months of monthly data, or 40 quarters of quarterly data) before positions are calculated, so each indicator’s history starts at a different point.

Where data have gaps, a gap of 1⁠–⁠2 periods is filled by linear interpolation between the periods on either side and noted on that indicator’s page (for example, during the US government shutdown in October 2025, the Bureau of Labor Statistics did not publish that month’s unemployment rate or prices). For longer gaps, the calculation starts again after the gap.

4. Global anchors, the four forces and the overall reading

From the 59 indicators of the first edition, this site picks 17 “global anchors”: representatives followed in every issue and shown on the dial on the home page. They are grouped into four forces: Money (the price and flow of money), Economy (growth, prices and their upstream drivers), Assets (the prices being pushed) and Amplifiers (sentiment, valuation and leverage).

The overall reading is the median of the positions of the 17 global anchors: half are hotter and half colder. It looks only at how far each indicator is from its own normal and in which direction, not at whether that “high” is good or bad for the economy. For example, a credit spread at the low end of its history (markets are not afraid of risk) pulls the overall reading toward the cold side, even though that usually happens when markets are optimistic. So the overall reading summarizes how far the big global indicators have swung from normal; it is not a score of how well the economy or markets are doing. To understand it, look further down at each force and each link. Each month counts only the anchors with data at the time (indicators are published at different times, so a position can be carried forward for up to 2 months), and at least 9 anchors must have data.

The four forces and 14 categories: indicators are combined into one position, again with the median, only when they point the same way (a higher reading means the same thing for all of them). Where directions are mixed (the Money and Amplifiers forces, and five categories: Central Banks and Money, Interest Rates, Currencies, Employment and Trade), combining them would let the extremes at either end cancel each other out, so they are not combined; instead, each indicator gets its own small chart. The Stock Markets category uses the median of 9 national and regional stock markets.

5. When it was here before: what happened next

For every position series, this site finds the past months at the same level as now (excluding the current period) and lists:

• Periods: months at the same level less than a year apart are merged into one period. For each period we give its start and end, how many of its months were at this level, and the cumulative change in US stocks (for price indicators, in their own price) 1, 2 and 3 years after it began. We do not measure from the last month: a period only ends once the indicator has stayed out of the level for 12 months, so that year is known in advance to have left the level, which would bias the figure toward a return to normal.
• 1, 3 and 5 years later: starting from every month at the same level, the median change, the share of declines, and the median deepest drop within 5 years, shown next to all months for comparison. Price indicators use their own price; the others use US stocks. “US stocks” means the S&P 500 price series compiled by Professor Shiller (monthly average, excluding dividends), from 1871, one of the longest stock price records in public data.
• Where it swung next: the median position 1, 3 and 5 years later, the share that returned to neutral (40⁠–⁠60), and the share still at the same level. The pendulum is only a metaphor: reaching an extreme does not mean it must swing back.

Neighboring months within the same market episode are not independent events, so later figures are listed only when there are at least 3 episodes and 12 months; otherwise we write “Sample too small to draw conclusions”. All of this is the historical record, not a forecast.

6. Has hot or cold lined up with later returns?

Every page sets out whether this yardstick is useful for that indicator. We compare the average annual return over the 5 years after its hottest 20% of months with the return after its coldest 20% (price indicators use their own price; the others use US stocks), and compute the rank correlation between position and later returns. Only when the absolute correlation is 0.25 or more do we write “Some link in the past”; otherwise we write “No clear link in the past”. With fewer than 60 months that have 5 years of later data, we write “Not enough data to judge yet”. The months overlap, so there are really only a few market episodes; this is a tendency, not a rule.

7. How links are tested

We collect commonly cited economic explanations of how indicators are linked (cause, knock-on, ripple) and test every link in three ways:

1. Common explanation: the usual economic explanation of the link, with a cited source (any still being checked are marked). This is the claim being tested, not a view held by this site.
2. Tested against our data: using the time lag given in the common explanation (for “0⁠–⁠12 months”, the midpoint of 6 months), we compute the correlation between the two positions (−1 to +1). We then look at every 5-year period, shifted one year at a time, and count how many move in the direction of the common explanation. Measured in the direction of the common explanation (with the sign flipped for opposite-direction links, called “agreement”), an agreement of 0.4 or more with at least 75% of periods in step is “supported by the data”; any other value above 0.1 is “weak or inconsistent”; between −0.3 and 0.1 is “not visible in our data”; below −0.3 is “our data show the opposite”. The agreement over the whole history and the result for each period can differ: two series can move together within every 5 years while their highs and lows do not line up over the whole history.
3. Periods when it did not hold: runs of periods that do not fit are listed in full.

When two indicators move together only because the same cause drives both (a common source: gold and the dollar, for example, both respond to real interest rates), we do not draw a link. Compounding and offsetting: common economic explanations describe some pairs as reinforcing or offsetting each other. We look only at the data: a pair lights up only when both indicators are at the named end (for example, US valuation at the high end and the credit spread also at the high end), and past periods with the same combination are listed.

8. Inclusion criteria

• Scope: indicators at the global level or for major economies.
• Durability: published by central banks, governments, international organizations, or researchers who maintain data over the long term.
• Length: ideally at least 30 years of monthly data; shorter series are still included, but their pages are marked “short sample”, and conclusions should be more cautious.
• Licensing: we use only data whose license allows public reuse, with attribution. Where a license does not allow it (for example, the VIX US stock volatility index, the DXY dollar index and purchasing managers’ indices), we use public data as a substitute and say so every time: they measure similar things, but they are not the same number.

9. Public substitutes: they measure similar things, but they are not the same number

10. Updates

Most indicators are monthly and are published in two waves: early in the month come last month’s share prices, exchange rates, commodities and the Shiller data; in mid-month come the OECD leading indicators, national share prices, inflation and BIS data. So this site publishes two issues a month (early and late in the month). Each issue automatically fetches the data, recalculates, writes its highlights and is archived. Every card shows its data month. If a data source changes format or a figure looks abnormal, updates pause and resume only after checking.

11. Data sources and licenses

• The World Bank: Commodity Price Data (The Pink Sheet); Global Economic Monitor; World Development Indicators: World Federation of Exchanges (stock market capitalization). License: CC BY 4.0; ratios and positions are calculated by this site and are not figures published by the World Bank.
• OECD: Main Economic Indicators (share prices), Composite Leading Indicators, business and consumer confidence, prices and quarterly national accounts, from the OECD Data Explorer (October 2026); attributed in line with the OECD Terms and Conditions. Positions are calculated by this site and are not figures published by the OECD.
• Bank for International Settlements (BIS): policy rates, effective exchange rates and residential property prices. The calculations and explanations on this site, in every language, are not official BIS versions.
• European Central Bank (ECB): the Composite Indicator of Systemic Stress (CISS), euro foreign exchange reference rates and long-term interest rates; positions are calculated by this site.
• Board of Governors of the Federal Reserve System (H.15, H.10, H.4.1, G.17, the Senior Loan Officer Opinion Survey, the Gilchrist–Zakrajšek spread), the Bureau of Labor Statistics and the Bureau of Economic Analysis, obtained via FRED from the Federal Reserve Bank of St. Louis; the Federal Reserve Bank of Philadelphia Manufacturing Business Outlook Survey (via FRED); the Federal Reserve Bank of New York Global Supply Chain Pressure Index; and the Financial Stress Index of the Office of Financial Research (OFR), US Department of the Treasury.
• Long-run US stock market data published by Professor Robert J. Shiller (shillerdata.com). S&P 500 is a trademark of S&P Dow Jones Indices LLC.
• Ministry of Finance, Japan, “Interest Rate” (JGBs), used under the Public Data License (PDL1.0); positions are calculated by this site from these data.
• Taiwan government open data: Central Bank of the Republic of China (Taiwan), 2026, “29.利率 A.中央銀行利率依期間” (central bank interest rates, by period; EG2AM01), “28.股票市場統計 B.股票交易與股價指數依期間” (stock trading and stock price indices, by period; EG27M01) and “我國與主要貿易對手通貨對美元之匯率依期間” (exchange rates of the NT dollar and major trading partners’ currencies against the US dollar, by period; BP01M01); National Development Council, 2026, “景氣指標及燈號” (business cycle indicators and signals); Department of Statistics, Ministry of Economic Affairs, 2026, “外銷訂單” (export orders); Directorate-General of Budget, Accounting and Statistics, Executive Yuan, 2026, “國民所得統計常用資料－季” (national income statistics, frequently used data, quarterly); Ministry of the Interior, Department of Land Administration, 2026, “115年第1季住宅價格指數發布內容” (housing price index release); all as downloaded in October 2026.
  此開放資料依政府資料開放授權條款 (Open Government Data License) 進行公眾釋出，使用者於遵守本條款各項規定之前提下，得利用之。
  Open Government Data License: https://data.gov.tw/license