Prof Scott Cederburg: Long-Horizon Losses in Stocks, Bonds, and Bills (EP.224)

Episode Summary

Executive Summary: Episode 224 features Professor Scott Cederberg discussing three practical finance problems: building a cleaner global return history, rethinking home-country bias and long-run asset risk, and choosing between pre-tax and post-tax savings accounts. His research suggests long-horizon real losses are more common than U.S.-only data imply, international stocks can hedge domestic inflation surprisingly well, and TFSA/Roth-style accounts can be valuable even for high earners because they eliminate major tax-rate uncertainty.

Main Topics: Constructing a global long-run return database (Priority: 5/5): Cederberg explains how he and coauthors built monthly return histories for 38 developed countries using GDP/industrialization-based development classifications, exchange-closure adjustments, and archival data filling via Google Translate and historical sources. Long-horizon equity risk and real loss probabilities (Priority: 5/5): The discussion centers on how domestic stocks can still deliver real losses over 30 years, why U.S.-only histories understate tail risk, and how much the left tail matters for retirement savers. International stocks, currency, and inflation hedging (Priority: 5/5): International equities are shown to have lower long-run real loss probabilities and near-zero correlation with domestic inflation because currency moves tend to offset local inflation shocks over longer horizons. Bonds and bills as poor inflation hedges (Priority: 4/5): Government bonds and bills are analyzed as long-horizon assets that often lose real purchasing power, with inflation—not default—driving most losses. Savings location: pre-tax vs post-tax accounts (Priority: 5/5): Cederberg’s tax-uncertainty paper argues that Roth/TFSA accounts can be valuable even for high-income investors because they lock in known tax treatment and reduce future tax schedule risk. Low beta anomaly vs conditional risk (Priority: 3/5): The low-beta premium is explained as an artifact of time-varying risk exposure; once beta is modeled conditionally, the apparent alpha largely disappears.

Key Arguments: U.S.-only historical returns are upward biased because they underrepresent survivor bias, market closures, and countries whose markets were destroyed by war, nationalization, or regime change. Domestic stocks in developed countries have a meaningful 30-year real loss probability; using the full dataset, the probability is about 13%, versus about 1% if only U.S. data are used. International stocks are more protective than domestic stocks over long horizons, with lower loss probability and a natural hedge against domestic inflation via currency depreciation when inflation rises locally. Currency hedging is useful at short horizons, but for long-term investors the unhedged exposure can be beneficial in real terms because FX changes offset inflation. Bonds and bills are not reliable long-run safe assets in real terms; their main risk is inflation eroding nominal yields over long horizons. The decision between pre-tax and post-tax accounts should incorporate uncertainty about future tax rates, not just current vs expected retirement income. Roth/TFSA-style accounts can be worth a substantial premium because they remove unrewarded tax risk; the paper finds investors may pay meaningful annual fees to optimize this choice. A simple heuristic from the tax paper is to hold pre-tax assets equal to age plus 20 percentage points, with the remainder in post-tax accounts. The apparent low-beta outperformance is largely explained by unconditional CAPM misspecification due to time-varying beta and varying market conditions. Saving more and maintaining flexibility in retirement withdrawals matter because even diversified portfolios can suffer meaningful long-run losses.

Data Points: Countries in dataset: 38 developed countries - Long-run return dataset used to study stocks, bonds, and bills Sample period: Starting in 1890 - Main dataset coverage for monthly returns Historical data volume: About 2,500 years of monthly return data - Combined panel across developed countries Domestic stock 30-year real loss probability: 13% - Estimated probability of a real loss for domestic stocks across developed countries U.S.-only 30-year real loss probability: About 1% - What the same analysis gives using only U.S. data International stock 30-year real loss probability: 4% - Estimated real loss probability for international stocks Domestic bond 30-year real loss probability: 27% - Long-horizon real loss probability for government bonds Domestic bill 30-year real loss probability: 37% - Long-horizon real loss probability for government bills Chance of losing at least half of buying power in stocks: 5% - Catastrophic-loss tail for domestic stocks over 30 years 1% tail for domestic stocks: About -80% real loss - Extreme downside in the 30-year distribution Chance of losing all money under 4% withdrawal rule: 17% - Married couple outcome using full developed-country sample Withdrawal rate for 95% success: 2.25% - Approximate safe withdrawal rate from the retirement withdrawal paper Annual fee willingness at 10-year horizon: Up to 7% per year - Value of optimizing account location under tax uncertainty Annual fee willingness at 30-year horizon: About 2% per year - Value of optimizing account location under tax uncertainty High-beta minus low-beta strategy alpha: -7% per year - Unconditional CAPM alpha found in low-beta literature discussion Conditional CAPM alpha for low-beta strategy: About 2% per year, statistically insignificant - After accounting for time-varying beta Monthly stock return sign balance: About 56% positive / 44% negative - Short-horizon domestic stock return distribution 5-year stock loss probability: 29% - Probability of negative return over a five-year holding period Domestic stock 90% spread over 30 years: 48 cents to $24 from a $1 starting investment - Middle 90% range of the 30-year distribution Loss probability if gaps/bias are reintroduced: Roughly cut in half - Effect of using cleaner biased data versus full corrected sample Probability all three domestic assets lose over 30 years: 6.5% - Domestic stocks, bonds, and bills all delivering real losses simultaneously

Pivotal Quotes: "we have about 2,500 years worth of monthly return data from these developed countries." — Scott Cederberg: Describing the scope of the long-run global return dataset "the probability of loss in half if you like introduced this bias back into the data" — Scott Cederberg: Explaining the effect of survivor bias and easy-data bias on estimated stock loss probabilities "just getting some money out of that system and into a post-tax account seems to have a really big economic benefit." — Scott Cederberg: Summarizing why Roth/TFSA-style accounts can be valuable even when pre-tax accounts look attractive

Implications: Investors should treat long-run equity, bond, and tax decisions as uncertainty-management problems, not just expected-return problems. Diversification across countries and account types can meaningfully reduce tail risk, while relying on U.S. history, bonds, or a single tax assumption can create false comfort.