The premise is routinely missed: Buffett is refusing to forecast macro swings, not refusing to study the rules that govern a business — he wrote insurance rate regulation, railroad tariffs and allowed utility returns into his own analysis. Fluctuation is noise; rules are parameters. The right response is not to guess the next policy document but to fold "the rules may change" into two places: a higher discount rate and a shorter credible forecast horizon. How many years you dare extrapolate depends on the stability of the rules, not on how good the company is.
The Shanghai Composite rose from roughly 2000 in mid-2014 to 5178 on June 12, 2015, with margin financing balances peaking near RMB 2.27 trillion; once deleveraging began, the index fell to 2638 by late January 2016, a drawdown of roughly 50%, during which IPOs were suspended and more than a thousand listed companies halted trading — liquidity itself became a policy variable. July 2021 went further: after the "double reduction" rules required academic tutoring firms to convert to non-profit status and barred capital-market financing, US-listed Chinese education companies broadly halved within two trading sessions, some falling more than 70% — this was not an earnings revision; the business model was withdrawn.
The opposite error is equally expensive: treating policy as the only variable degrades investing into guessing documents and chasing themes. Policy also creates value — opening a sector to entry and industrial support are real sources of profit. The boundary is this: policy risk does not mean uninvestable; it means you have to buy it cheaper. A persistent discount may itself be the compensation. The only questions are whether the compensation is large enough, and whether you can absorb the permanent loss in the worst case.
China's A-share market stretches the time scale of that sentence. Individual investors hold roughly a fifth of the free-float market value yet account for around 80% of turnover — prices are set by participants whose holding period is measured in weeks, while fundamentals move in years. The mismatch manufactures durable theme premia and amplified sentiment: greater volatility, but mispricings that also persist longer.
The 2015 margin book was this structure crossed with leverage at its most extreme: forced-liquidation selling and sentiment feedback reinforced each other into a self-fulfilling decline. The mirror image came in 2020–2021, when money flowed through mutual funds into a small set of "core assets," pushing their valuations to highs before a steep drawdown after the 2021 Lunar New Year — institutionalization is not automatically rationality; it swaps a retail herd for an institutional huddle.
Three limits. The structure is changing: A-shares entered the MSCI indices in 2018 and the foreign and mutual-fund share has risen, so describing today with decade-old turnover figures distorts. "Lots of retail, therefore inefficient" is a lazy inference: heavily covered large caps are often priced perfectly well, and the inefficiency concentrates in thinly covered corners — which are also where information quality is worst. The most dangerous error is mistaking structural inefficiency for personal skill: exploiting it takes capital duration and psychological endurance, not superior judgment.
State-owned enterprises carry mandates for employment, price stability and supply security, and those mandates show up in the accounts as suppressed margins and elevated capital expenditure — not incompetence, but the objective function at work. The valuation discount is therefore often rational rather than mistaken: the same free cash flow deserves a discount if part of it will be spent on non-shareholder purposes in bad years. Conversely, their regulated, utility-like cash flows can suit capital that wants duration and dividends. A private firm's objective function is purer, at the cost of founder-concentrated governance, weaker policy bargaining power, and a higher cost of capital.
Coal-fired power in the autumn of 2021 is the demonstration: thermal coal prices spiked while on-grid tariffs stayed regulated, so generators ran at widespread losses even as some provinces rationed electricity — costs marketized, prices administered, the spread absorbed by the company. On the valuation side, listed A-share banks have traded below book value for years; in November 2022 regulators called for exploring "a valuation system with Chinese characteristics," and state-owned sectors subsequently re-rated — evidence that the discount is neither purely an error nor permanent. Governance risk, meanwhile, respects no ownership type: Luckin Coffee disclosed roughly RMB 2.2 billion of fabricated sales in April 2020 and was subsequently delisted from Nasdaq.
"State-owned means inefficient, private means efficient" is an oversimplification: the cash-flow quality of a monopoly-adjacent state firm often beats that of a private firm in a fiercely competitive industry. "The state will bail it out" is more dangerous still: an implicit guarantee protects creditors, not shareholders, and recapitalizations frequently arrive through dilution — treating state support as a floor under the share price applies bond logic to equity. And the discount can be a value trap: paired with an ROIC persistently below the cost of capital, cheap is only cheap on paper.
Most US-listed Chinese shares are shares in a Cayman entity that captures the economics of an onshore operating company through a chain of contracts — the variable interest entity (VIE) structure. Three risks stack on that chain: structural risk (the contracts work as long as regulators tolerate them), listing-venue risk (cross-border audit oversight and delisting rules), and transferability risk (foreign-exchange and capital-flow management). What they share is that they are independent of business quality: the company can execute flawlessly and the certificate can still lose its trading venue. A margin of safety cannot defend against this, because it does not act on the cash flow — it acts on the link between you and the cash flow.
On November 3, 2020, Ant Group's listing — due the next day, raising roughly USD 34.5 billion, then the largest IPO ever — was suspended. A ride-hailing platform listed on the NYSE on June 30, 2021, was placed under cybersecurity review two days later with its apps removed from stores, and delisted about a year afterwards: what investors faced was not an earnings swing but the disappearance of the trading venue itself. The US Holding Foreign Companies Accountable Act became law in December 2020, briefly putting the entire China ADR complex at risk of delisting; in August 2022 US and Chinese regulators signed an audit-oversight agreement, and that December the PCAOB announced it had secured complete inspection access, materially easing the risk.
The symmetric side has to be said. A tail risk is not a certainty: the VIE structure has existed for over two decades without being invalidated wholesale, and the worst case of mass delisting never materialized. Reasoning backwards from the worst case to "never touch it" is itself expensive — some assets priced for despair in 2022 delivered handsomely afterwards. Both extremes are wrong: pretending the structural risk does not exist (holding these names with the position discipline you would use for US equities), or treating it as uninvestable. The middle answer is a position cap you can live with, plus falsification conditions written down in advance. One more shortcut to guard against: economic growth is not shareholder return — Ritter (2005), using over a century of data across 16 countries, found the cross-sectional correlation between real per-capita GDP growth and real equity returns to be negative.