Research
Industry views & methodology
General research written for all readers. We talk about method and data — not entry/exit calls or hot tips.
Running Hot, or a Moving Ruler? GDP, AI, and Whether 'Normal' Inflation and Interest Rates Change With the Times
On September 30 the BEA publishes its third estimate of second-quarter GDP. Two stories are circulating ahead of it: the economy is running too hot and the Fed will keep hiking; and AI is opening a new era in which the old yardsticks no longer apply. This article tests both against public data. What is high is prices, not output. AI investment has reached a record share of GDP going back to 1947, while an AI productivity boom is not yet visible. The yardstick for interest rates really does move with the times; the yardstick for inflation is deliberately held still. And the great technological breaks of the past (the Industrial Revolution, electricity, the internet) changed the ruler for growth, not the ruler for inflation.
Read articleIs 5% High? US Treasury Yields Since 1962, and the Ruler That Decides the Answer
The 10-year Treasury yield closed at its highest level since October 2023, and the 30-year at its highest since July 2007. Whether 5% is high depends entirely on which stretch of history you measure it against: against every trading day since 1962 today's level sits below the median, while since 2008 only 0.19% of days have been higher. This article measures that ruler from public data, and takes apart why rates reached 15.84% in 1981, what pulled them down for forty years, why they came back after 2021, and what presidents, oil, mortgages and the technology industry each had to do with it.
Read articleWhat We Call “Working” Is Not What You Call “Making Money”
The first two articles tested 20,934 chart patterns and 49 technical signal types. Both returned zero. But readers keep asking the same question: then why do some people make money? Half the answer is not in the data — it is in the definitions. We never measured profit, our shortest holding period is one trading day, and we require a signal to survive an average of 2,661 occurrences while a trader needs only their own few dozen trades. Change nothing but the benchmark and the share of test cells that "work" falls from 81% to 22%.
Read article49 Technical Signals, Zero Passed
Moving averages, stochastics, RSI, MACD, Bollinger Bands, candlestick patterns, Fibonacci retracements, plus multi-timeframe trend conditions — 49 signal types, 1.78 million events, 11,025 test cells, and zero survived FDR correction. But the real story is this: following my own written specification produced 5 significant findings, four of them Fibonacci. What stopped them was a control run I did not strictly need — because the control produced 5 as well.
Read article20,934 Chart Patterns, Zero Passed
A point-in-time reconstruction of S&P 500 membership from 2000 to 2017, 20,934 detected chart patterns, three layers of benchmark, and a multiple-testing correction. Of 245 test cells with adequate sample size, zero survived. The scope is chart patterns only — RSI, MACD, stochastics and the rest were never tested. This article is about that zero, about an intermediate conclusion I had to retract, and about a question the study cannot answer: even if you do beat the index, is the excess worth what it costs you.
Read articleWhat It Took to Collect the Average: Crashes, Recovery Time, and the Cost of Being Wrong
US equities returned about 6.9% a year in real terms over 126 years. This is the bill for that number: every drawdown of 20% or more, how long each took to recover, why the answer changes completely depending on which ruler you use, and the arithmetic of how much a side bet actually costs when it fails.
Read articleThe Break-Even Line: 126 Years of US Inflation and What Standing Still Costs
Burton Malkiel's point that merely breaking even requires a return equal to inflation is easy to nod along to and harder to feel. So we measured it: every year of US inflation from 1900 to 2025, what cash and bonds and stocks actually returned against it, and why the single most-quoted number about the stock market leaves out most of the return.
Read articleHow to Calculate the S&P 500 Index from Individual Stocks
A step-by-step walkthrough of the divisor-based, float-adjusted market-cap formula behind the S&P 500 — how to turn one stock's price move into index points, where every input actually comes from, and exactly where the common shortcuts break down.
Read article