Independent market analysis
Long-form research on portfolio construction, risk, macro and the tools that actually move returns. No hype, no tips, no paywall.
From 1928 to 2025 the S&P 500 averaged 11.85% a year but compounded at 10.02%. The gap, roughly half the variance of returns, is volatility drag. It is why about two in three simulated 30-year histories fall short of an average-return projection, and why a simulated 3x daily fund lost 52% from 2000 to 2007 while the index gained 12%.
On Shiller's CAPE, the S&P 500's earnings yield (2.44%) fell below the 10-year TIPS real yield (2.85%) on 24 September 2026, the first negative gap since at least 2003; on FactSet's forward P/E the gap is still about 2.4 points, roughly half its ten-year norm. Nearly all of the squeeze since 2021 came from higher real yields, not higher valuations.
In December 2023 Argentine prices rose 25.5% in a single month. In August 2026 they rose 1.7%. Two consecutive years of fiscal surplus, three rating upgrades between May and July, record oil output and a trade surplus five times larger than a year earlier took country risk from roughly 1,900 basis points to about 400 at its July low. The bond market has paid for most of that rebuild. Argentine growth — wages, consumption, earnings — has not been repriced yet, and that is the half still available.
Over the 20 years to June 2026, 92.6% of US large-cap active funds trailed the S&P 500 after fees, and only about 51 of 690 both survived and beat it. The harder problem is picking them in advance: five-year persistence among winners was no better than chance.
The 10-year Treasury closed at 5.01% on 16 and 18 September 2026, the highest since July 2007, after the Fed's first hike in more than three years. Since 1 May the whole rise has been in real yields: breakeven inflation fell from 2.48% to 2.33%, while consumers now expect 4.6% inflation next year.
The S&P 500 has had eleven bear markets since 1950. The median one took 24.5 months to regain its previous peak in price terms — but price ignores dividends and inflation. Count both, and the 2000 peak was not recovered until May 2013: one drawdown of nearly thirteen years, not two separate crashes.
Six biases measured on real accounts: trading frequency cost the most active US households 7.1 points a year, the disposition effect 3.4, switching holdings 3.2, and fund timing 1.2. The numbers are real, they come from different samples and decades, and adding them together — as most articles do — produces nonsense.
Backtrader tops almost every 2026 ranking of Python backtesting libraries. Its last commit was in April 2023. Here is what the registries say about all six, and the licence question nobody mentions.
The Kelly criterion says bet 10% per trade on a 55/45 edge — the growth-optimal fraction. Simulated over 200 trades, that same fraction produces a 50% peak-to-trough drawdown in 93% of runs, and ends below half the starting capital once in ten. Half Kelly keeps three-quarters of the growth and a fraction of the pain.
Run $10,000 plus $500 a month at 7% for 40 years and the calculator shows $1.48 million. Only 17% of that — $250,000 — is money you actually put in. The other 83% is growth compounding on growth, and by the final decade it does nearly all the work.
