Foundation & Philosophy

For four decades a simple portfolio worked. Ten-year Treasury yields fell from above 15 percent in September 1981 to 0.62 percent in July 2020, so bonds paid income and gained in price at the same time. Inflation stayed contained, and from about 2000 bonds also tended to rise when stocks fell. A 60/40 mix of US stocks and 10-year Treasuries returned about 7.8 percent a year after inflation from 1982 to 2021. Much of that reputation was earned by falling rates, and repeating that tailwind would take a starting yield of 15 percent.

The 2020s run on different terms. Inflation has returned as a separate source of uncertainty: in 2026 the annual rise in consumer prices went from 2.4 percent in January to above 4 percent in May, and was 3.4 percent in August. Central banks now face a conflict they had not faced in forty years. A year after the Federal Reserve began raising rates from near zero, losses on bank bond portfolios helped set off the run that closed Silicon Valley Bank on March 10, 2023. Two days later the Fed announced a facility lending to banks against Treasuries, agency debt and mortgage-backed securities valued at par rather than at market. It kept raising rates that spring, but only behind a new backstop. Fighting inflation and protecting the financial system can now pull in opposite directions.

The post–Cold War order is splitting into competing blocs, and supply chains, reserves and payment systems are splitting with it. Technology pushes the prices of many goods down while asset prices climb, so the consumer price index and the cost of owning assets drift apart. Markets also show volatility clustering: calm and storm arrive in runs.

Judged against consumer prices alone, a portfolio can keep pace with inflation and still fall behind the assets that store wealth. Consumer prices rose about 88 percent from the end of 1999 to the end of 2024. Home prices more than tripled. An equal mix of stocks (with dividends), homes, and gold rose about sixfold. Keeping pace with the cost of living turned out to be a low bar.

Not all of that gap is monetary: real growth and falling interest rates account for part of it. Falling rates also hid the cost of debt. From 1991 to 2021, federal debt held by the public rose from about 44 to 94 percent of GDP while the interest bill fell from 3.2 to 1.5 percent of GDP. Then rates rose. By fiscal 2025 the debt stood at 98 percent of GDP on the OMB series plotted below, and the bill was back at about 3.2 percent, level with its 1991 peak, on more than twice the debt.

The arithmetic now runs in a loop. Deficits require new borrowing, new borrowing at higher rates raises the interest bill, and the interest bill widens the deficit. Under current law, the Congressional Budget Office projects that debt held by the public will pass its 1946 record of 106 percent of GDP around 2030 and reach 120 percent by 2036, with net interest rising to 4.6 percent of GDP (February 2026 baseline). When a government’s debt and the cost of carrying it begin to limit what its central bank can do, economists call the condition fiscal dominance; Thomas Sargent and Neil Wallace set out its arithmetic in 1981. We treat it as a governing constraint of this regime, though it has not clearly bound yet. From August 2023 to November 2025 the Federal Reserve held its policy rate at least a point above core inflation while the interest bill climbed past 3 percent of GDP. It then cut three times between September and December 2025, held through the summer of 2026 while core inflation rose back to about 3 percent, and on September 16, 2026 raised its rate for the first time since 2023. A central bank raising rates on top of a 3 percent interest bill is a point against our reading; the first test at the end of this Part says what would show that the constraint is not binding.

The United States has carried a debt this size once before, and how it came down is the most useful precedent we have. Debt held by the public peaked at 106 percent of GDP in 1946 and fell below a quarter of GDP by 1974, although in dollars it was larger in 1974 than in 1946. It was never paid off. From 1942 to 1951 the Federal Reserve held Treasury yields down at the Treasury’s request, with an implicit cap of 2.5 percent on long-term bonds, while consumer prices rose by more than a third between the end of 1945 and the end of 1950. Carmen Reinhart and M. Belen Sbrancia (NBER, 2011) call the result the liquidation of government debt: across the advanced economies they studied, real interest rates were negative roughly half the time between 1945 and 1980, and for the United States and the United Kingdom they estimate the liquidation averaged 3 to 4 percent of GDP a year. Growth alone would not have done it. Julien Acalin and Laurence Ball (NBER, 2023) attribute 51 of the 83 points of decline to primary budget surpluses and distorted interest rates together; without them, the ratio would have fallen only to about 74 percent of GDP. Bondholders carried much of the cost: in Damodaran’s series, 10-year Treasuries lost about 1 percent a year after inflation from 1946 to 1974, while a 60/40 mix still returned about 3.3 percent, carried by its stocks. This time the budget has no surplus to share the work: the CBO baseline has a primary deficit in every year through 2036, 2.6 percent of GDP in 2026 and 2.1 percent in 2036. A portfolio that relies on government bonds for its real return is betting that this debt will be resolved more kindly to bondholders than the last one was.

Fragility is not volatility. Volatility is the size of the swings; fragility is the inability to survive them, and a single volatility figure measures only the first. A portfolio that drops 15 percent in a garden-variety correction and recovers is volatile but sound. A portfolio that takes a permanent loss because its diversifiers fall at the same moment, in a correlation breakdown, is fragile. An average volatility figure cannot tell the two apart.

The hedge most portfolios rely on belonged to one period. Through the 1970s, 1980s and 1990s, US stocks and Treasuries usually moved together. From about 2000 to 2021 they tended to move in opposite directions, so bonds cushioned equity losses. In 2021 and 2022 the correlation turned positive again. Research at AQR (2023) offers an explanation: stocks and bonds respond in opposite directions to news about growth and in the same direction to news about inflation, so the sign of their correlation depends on which kind of uncertainty dominates. Their model explains about 70 percent of the long-term variation in the US correlation.

In 2022 a 60/40 mix of US stocks and bonds fell about 16 percent: stocks lost about 18 percent with dividends, and bonds, the part meant to cushion them, lost about 13 percent. With 10-year Treasuries as the bond half it was worse: in Damodaran’s series the Treasuries lost 17.8 percent, and 2022 is the only year since 1928 in which US stocks and 10-year Treasuries both fell by more than 10 percent. After inflation the mix lost about 23 percent, a loss matched or exceeded only in 1937 and 1974. In 2022 bondholders lost because yields rose; in the 1940s they lost because yields were held down. The second mechanism is the one this thesis expects. For a retiree drawing 4 percent a year, a loss of that size early in retirement does lasting damage, because withdrawals taken from a smaller balance leave less capital to recover with. That is sequence-of-returns risk.

The problem was not bad luck. The problem was not “this time was different.” The problem was that the portfolio was not designed for the regime it inhabited.

The same years tested a very different holding. Bitcoin fell more than 70 percent twice, in 2018 and again in 2021 and 2022; the second fall, from its November 2021 high to its November 2022 low, was about 77 percent, and in calendar 2022, with inflation at a forty-year high, it lost about 64 percent. Over short spans it has traded like the riskiest of risk assets. A fixed purchase every month from January 2018 to December 2024, made through both falls, still ended at about seven times the money put in. That result requires a position small enough that nothing forces a sale at the bottom, and it says nothing about the next seven years.

Hindsight makes these results look obvious, and they do not show that regime calls can be made reliably in advance. No one knows the timing of a regime change, including us. The framework therefore spends as much effort on surviving a wrong call as on making a right one, and it starts from a different premise:

A forecast names an outcome and a date. A wrong forecast is expensive, and one that gets the direction right can still miss on timing, which decides whether a position survives long enough for its thesis to be proved.

Timing matters most once withdrawals begin, because they make the order of returns count. Take $1 million, twelve years of returns and $40,000 withdrawn at the end of every year, and run it twice: once with the gains first, once with the same returns in reverse.

Survivable compounding is our name for the objective: a portfolio that can be wrong about timing, size or specific outcomes and remain intact, while keeping enough exposure to convex payoffs, where being right pays far more than being wrong costs. With payoffs of that shape, bought at a sensible price, being right only some of the time can compound to far more than a risk-free return.

That shape has a cost. Near the middle, the curved line sits a little below a plain straight-line bet; that gap is the price of convexity. Sizing keeps the left side survivable, and the right side is left open, within the concentration limits.

We built the framework to be audited: every decision should trace to a written rule or a recorded judgment that the investor could explain to an advisor, a client or an heir. It has seven components, each aimed at a known way traditional portfolios fail:

The framework holds Bitcoin for what it does. It is a long-duration monetary asset that runs parallel to the dollar system, with a supply capped at 21 million coins by its protocol and growing institutional access. In a regime whose central risk is the erosion described above, an asset whose supply no government can expand is, in our view, the most direct answer to that risk. It is also among the most volatile assets an investor can own. It is scored like every other holding, but sized and handled under its own rules, and it is held whatever the thesis. Part 3 makes the full case and sets out the four valuation models that pace its buying: power-law bands, realized price, network adoption and production cost. A fifth reading, global liquidity, gives direction but sets no value.

Account placement matters most for strategies whose returns come from a few large wins. Over decades, where a convex holding sits (Roth, taxable, or pre-tax) can change the after-tax result by a wide margin; Parts 4 and 5 work the arithmetic and state their assumptions. At equal tax rates in and out, Roth and pre-tax end even. For convex, right-tail positions the balance tips toward Roth because the same dollar limit shelters more after-tax money in a Roth than in a pre-tax account, required distributions can push large pre-tax balances into higher brackets, and heirs generally inherit a Roth free of income tax under current law (Part 4).

For an investor building wealth for a family, the framework is a repeatable process; for an advisor, a documented alternative to fee-for-beta management, though not a substitute for a suitability process. Its case rests on a thesis that can be checked: the last section of this Part sets out the evidence that would prove it wrong.

The order matters because each stage feeds the next. Positions bought without a macro thesis are exposure with no reason behind them, and a thesis with no tripwires tends to break in silence, so the investor hears about it last.

Six failure modes recur, each with a named answer:

A governed process takes discretion away where it does damage (emotional sizing, holding on a narrative, ignoring tax) and leaves it where we think judgment can add value (thematic rotation, tactical positioning and adaptive use of borrowing).

The investor still makes every decision; the written rules mean the hardest ones were thought through on a calm day, before the selloff that tests them.

Survival comes first because only capital that survives can keep compounding.

These are the tests we have set for the thesis of this Part. That thesis is our reading of the structural regime; the macro thesis that sets a portfolio’s structure belongs to the practitioner (Part 2), and the method does not change with either. Every test counts only from January 1, 2026: what happened before then is reported as history and does not count toward meeting a test. About nine months of that window had passed when we set them. Each test has public data, a window, a threshold set by our judgment with the history in view, and a stated consequence. A test that is not met is not evidence for us: most of these outcomes have causes other than the one we argue. The publication date, the release commit and an Internet Archive snapshot of this wording will be stamped here at release; if we ever change a test, we will say so and keep reporting against the original.

The exact method, the data and the history behind each threshold are on the Evidence page, with the script that computes the readings. We will publish the readings each February, including the years when nothing changes. These tests check the thesis, not the framework: whether following the framework does better than a simple portfolio after costs and taxes is a separate question that nothing here answers yet. A single bad year proves nothing in either direction, and neither does a Bitcoin drawdown; Part 3 judges the backbone against ten requirements and lists what could break it.