We've built an elaborate machine for processing beliefs about the future, but have yet to make any of said beliefs about the future. This is a problem. The preference framework from The Geometry of Preference (measure the left tail with CVaR, penalize it with λ, find the tangent point) and isolated the 4 forward-looking factors for capital asset pricing from Ending Capital Asset Pricing's Geocentric Era requires an input I haven't yet constructed: the forward-looking distribution itself.
We still need beliefs about what returns will look like (not what they looked like) and this article provides those beliefs through what else: AI.
Historical returns won't do for the identical reason so much of the models rethought in this blog don't: you can't extrapolate from descriptive data. The world that generated 1926-2024 returns—Bretton Woods, the Cold War, Pax Americana, the Great Moderation, zero interest rate policy—is not the world that will generate 2026-2032 returns.
Point forecasts won't do either. A single estimate of "expected return" hides the uncertainty that matters most. Two forecasters might both predict 7% equity returns, but one believes the range is 5-9% and the other believes it's -10% to +25%. These are radically different distributions with radically different portfolio implications.
The alternative is a structured way to enumerate possible futures, assign probabilities to each, and derive return distributions from explicit beliefs about the world. Not because anyone can predict the future—but because explicit uncertainty is more useful than implicit assumptions.
Correlation Isn't a Number
Extrapolating historical correlation is another instance of confusing descriptive statistics with causal structure. The correlation between any two assets depends on which macro forces are active—and different scenarios activate different forces.
Equities and Treasuries showed negative correlation for twenty years—around -0.3 to -0.4 through the 2000s and 2010s. If you're building a portfolio, you might look at that number and think you understand something about how these assets move together. The entire 60/40 portfolio is built on this relationship: when stocks fall, bonds rise, cushioning the blow. But that negative correlation was an artifact of a specific macro regime: disinflation, falling rates, the "Fed put." Growth shocks hurt equities but triggered flight-to-safety flows into Treasuries. Rate cuts to rescue markets boosted bond prices. The world was configured to make them move opposite.
Change the configuration. Inflation returns, central banks tighten aggressively, real rates rise from negative to positive. Now both equities and Treasuries are hurt by the same force—rising rates crush bond prices while simultaneously compressing equity multiples. In 2022, the correlation flipped positive. The same two assets that "moved opposite" now moved together—not because something changed about the assets, but because the macro forces acting on them changed.
In that sense, there isn't a causal connection between Stock A and Stock B that leads to a positive correlation, but a common cause upstream of both securities that produces correlated movements only as long as it remains active. Change the upstream cause, or activate a different one, and the downstream correlation changes with it.
Correlation emerges from common causes, not direct links between securities.
The mistake is treating correlation as an input when it's actually an output. This sounds like a minor technical point but it's really the whole thing. A strategy that treats correlation as a point estimate will break down at the worst possible time. Correlation emerges from shared exposure to macro forces. Change the macro configuration, change the correlation. A single historical number hides this conditionality—and hiding it means you can't anticipate when your diversification will fail precisely when you need it most.
The best method for getting to this distribution, I think, is to map out expected returns from scenarios based on first principles' analysis. Not to predict the future, but to understand the conditional structure of asset relationships. If you know why two assets are correlated—which macro exposures they share—you can reason about when that correlation will hold and when it will break. The goal isn't a better point estimate of correlation. It's a distribution of correlations across possible worlds, each weighted by probability.
Exogenous Forces
In this article I'll isolate the exogenous factors introduced in Ending the Capital Asset Pricing Geocentric Era—the macro forces outside any firm's control that nonetheless shape returns. To avoid contamination from idiosyncratic factors, I'll focus on liquid ETFs where the time horizon and ensemble diversity wash out the endogenous, push, and reflexive components of price fluctuations (i.e., exogenous forces comprise the vast majority of price fluctuations).
These exogenous forces define the game board on which companies compete. A semiconductor company can optimize its operations, but it can't control whether the US and China decouple. A utility can manage its grid, but it can't control whether carbon pricing becomes law. A bank can hedge its positions, but it can't control whether rates stay higher for longer. Interest rates, geopolitical alignments, regulatory regimes, technological trajectories, demographic shifts—these deep parameters determine which strategies succeed and which fail.
The key insight: if you can enumerate the major exogenous forces and their plausible outcomes, you can construct scenarios that represent coherent possible worlds. Each scenario implies different return distributions for different securities. Weight the scenarios by probability, and you have a forward-looking distribution built from beliefs rather than backward-looking statistics.
13 Questions
After extensive analysis of what actually moves markets over decade-long horizons, I've distilled the exogenous landscape into thirteen questions. These aren't predictions—they're the axes along which the future will vary. Each question has four possible outcomes. Different combinations of outcomes produce different worlds, and different worlds produce different returns.
Below are all 13 macro regimes and 4 distinct and complete scenarios (i.e., one scenario, no more or less, must come to pass) pertaining to each regime. The probabilities shown in parentheses are baseline weighted probabilities coming from a diverse array of humans and LLMs.
Each question represents an axis along which the future can vary. Four outcomes per question yields 52 distinct outcome states. Outcomes within a question are mutually exclusive; outcomes across questions can combine.
The thirteen questions are chosen to span distinct macro dimensions—US politics, China trajectory, AI progress, energy transition, and so on. They're also chosen for impact on asset returns over the completely arbitrarily chosen 6 year time horizon (actually, I somewhat intentionally chose this cutoff to avoid having to predict 2 U.S. elections).
From Questions to Scenarios
Each question has four mutually exclusive outcomes. Four outcomes across thirteen questions yields over 67 million possible combinations, which is why we don't enumerate all of them. Instead, I combine outcomes into coherent scenarios—internally consistent worlds where the answers to different questions reinforce each other. A "Constitutional Crisis" combined with "Dollar Downturn" is more coherent than "Constitutional Crisis" with "Permanent Dollar." A "Green Tech Acceleration" world is more likely to coincide with "Resilience Victory" on climate than with "Security First" energy policy.
Figure 1 Example Combined Scenarios
Four example combined scenarios showing how outcome states across questions form coherent worlds. Each tag represents one outcome from one of the 13 questions. Probabilities are illustrative—your own assessment may differ. The full analysis requires assigning probabilities to each of the 52 outcome states and understanding their conditional dependencies.
Six Scenario Packages
Beyond the illustrative combinations above, I've constructed six coherent scenario packages that represent distinct macro regimes. Each package sets one or more outcome states to 100% probability while other questions maintain their baseline probabilities.
Thirty Securities
With scenarios defined and weighted, the next question is: how do different parts of the market respond to different scenarios?
I use 30 securities spanning five asset classes, each investable through liquid ETFs:
US Equity (11 securities): The 11 GICS sectors via Fidelity MSCI ETFs—FCOM (Communication Services), FDIS (Consumer Discretionary), FSTA (Consumer Staples), FENY (Energy), FNCL (Financials), FHLC (Healthcare), FIDU (Industrials), FTEC (Information Technology), FMAT (Materials), FREL (Real Estate), FUTY (Utilities).
Developed Markets Equity (6 securities): SPDR international sector ETFs where they exist—IPK (Technology), IPF (Financials), IRY (Healthcare), IPS (Consumer Staples), IPD (Consumer Discretionary), RWX (Real Estate). DM offers sector granularity but not full 11-sector coverage.
Emerging Markets Equity (5 securities): Country and regional ETFs rather than sectors—IEMG (Broad EM), FXI (China), INDA (India), EWT (Taiwan), EWZ (Brazil). EM sector ETFs are illiquid or nonexistent; country exposure captures the macro sensitivities that matter.
Fixed Income (6 securities): The duration and credit spectrum—TLT (Long Treasuries 20+ yr), IEF (Intermediate Treasuries 7-10 yr), LQD (Investment Grade Corporate), HYG (High Yield Corporate), TIP (TIPS/Inflation-Linked), EMB (EM Bonds USD).
Alternatives (2 securities): Real asset exposure—GLD (Gold), DBC (Broad Commodities).
The asset class structure matters because different macro forces drive distinct changes to the market capitalization of each class. Equities generally suffer in crisis scenarios. But long Treasuries often rise during equity crises—flight to safety pushes yields down and prices up. Gold rises during uncertainty and inflation. High-yield bonds look like equities with credit risk layered on top. These structural differences create genuine diversification that sector rotation within equities cannot provide.
Each security has a sensitivity profile: how its returns respond to each of the thirteen macro questions. FTEC is highly sensitive to AI trajectory and interest rates. TLT is highly sensitive to inflation and rate policy—but with opposite sign to equities. GLD responds to dollar weakness, inflation fears, and geopolitical uncertainty. EMB combines EM equity sensitivity with duration risk.
Figure 2 Security Sensitivity Matrix (Partial)
Sensitivity matrix showing how strongly each security responds to each macro question. Red indicates high sensitivity; yellow indicates low. The full matrix covers all 30 securities × 13 questions. For a given scenario, you know the macro outcomes. You can trace through the matrix to estimate each security's return in that scenario—accounting for the structural differences between equities (which generally suffer in crisis), Treasuries (which often rally), gold (which rises on uncertainty), and credit (which behaves like equity with spread risk).
The sensitivity matrix is the bridge between scenarios and security returns. For a given scenario, you know the macro outcomes (US-China = cold war, inflation = persistent, etc.). You can trace through the matrix to estimate each security's return in that scenario. Aggregate across probability-weighted scenarios, and you have a forward-looking return distribution for each security.
Deriving Correlation
Earlier I argued that correlation isn't a number—it's conditional on the macro configuration. The scenario framework makes deriving it straightforward.
Within each scenario, securities that share positive macro exposures will co-move positively; securities with opposing exposures will co-move negatively. The scenario framework makes these exposures explicit through the sensitivity matrix. For US Technology and EM Technology:
Figure 3 How Correlation Emerges from Shared Exposures
The same sector across different geographies shows different correlations under different scenarios. US and EM Technology are highly correlated in benign environments but diverge sharply when US-China tensions escalate. Understanding the macro drivers of correlation lets you anticipate when geographic diversification will and won't work.
The realized correlation is the probability-weighted average across scenarios—but that aggregate number is less useful than understanding the distribution. A portfolio optimized for ρ = +0.4 will behave very differently depending on whether that average comes from consistent +0.4 across scenarios or from +0.8 in some worlds and -0.5 in others. The scenario framework preserves this structure.
This is what broke portfolios in 2022. The historical stock-bond correlation of -0.3 was an average across a specific macro regime. When the regime shifted to inflation, the correlation flipped to +0.4. Investors who understood correlation as derived from macro exposures—both stocks and bonds are duration-sensitive assets that suffer when rates rise—could anticipate this. Investors who treated -0.3 as a stable parameter were blindsided.
The inclusion of fixed income and alternatives fundamentally changes the correlation structure. In the simulations:
TLT correlates at -0.26 with FTEC: When tech crashes, Treasuries rally. This isn't diversification theater—it's structural. Both respond to rate expectations, but with opposite signs.
GLD correlates at -0.33 with FDIS: Gold rises when consumer discretionary falls. Crisis scenarios push capital toward real assets.
HYG correlates at +0.47 with IEMG: High-yield bonds behave like equity with credit risk. They're not "bonds" in the diversification sense.
TLT correlates at -0.31 with HYG: The flight-to-safety trade—when credit spreads blow out, Treasuries rally.
These correlations aren't stable numbers. They're scenario-dependent. In the "Stagflation Redux" scenario, TLT and equities become positively correlated—both suffer from rate hikes. In "Constitutional Crisis," TLT and GLD both rally while equities crash. The framework preserves this conditionality.
Eliciting Real Beliefs
The probabilities throughout this article come from a hybrid forecasting panel: six humans and six AI models, each prompted through a two-stage process adapted from Philip Tetlock's superforecasting research. Stage one forces the outside view—historical base rates, reference class frequencies, what the literature says about similar transitions. Stage two allows the inside view—specific causal mechanisms, current trends, necessary conditions. The structure prevents narrative reasoning from overwhelming base rates. Weighting the results across forecasters produces the consensus probabilities. No individual forecaster dominates; disagreement is preserved in the variance.
We asked AI models to forecast AI's trajectory. They were not pessimistic. Rimas, a human, assigned 70% to AI progress plateauing, a sentiment echoed somewhat less expressively, but still strong, by most humans, while the AIs assigned ~15%. Make of that what you will.
Figure 4 Forecaster Responses (Q1–Q4)
| Outcome | GPT 5 Pro | Sonnet 4 | Gemini 2.5 | Opus 4 | Deep Seek | Grok 4 |
Avg Legacy |
Sam | Avg Hum Ex-Sam |
Simple Avg |
Wtd Avg |
σ |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Q1: US Political & Social Contract | ||||||||||||
| Constitutional Crisis | 11 | 9 | 30 | 18 | 3 | 20 | 17 | 18 | 24 | 17% | 16% | 7.3 |
| Democratic Restoration | 23 | 34 | 28 | 28 | 23 | 7 | 26 | 27 | 24 | 24% | 24% | 7.0 |
| Populist Mandate | 28 | 20 | 19 | 40 | 27 | 17 | 31 | 15 | 17 | 24% | 24% | 7.8 |
| Gridlock Equilibrium | 38 | 37 | 23 | 14 | 47 | 56 | 26 | 40 | 34 | 35% | 35% | 12.5 |
| Q2: Europe: Power, Defense & Cohesion | ||||||||||||
| Defense Supercycle | 18 | 23 | 38 | 28 | 20 | 18 | 25 | 16 | 21 | 23% | 22% | 6.5 |
| Illiberal Equilibrium | 26 | 23 | 42 | 22 | 25 | 22 | 26 | 13 | 18 | 24% | 24% | 7.5 |
| Technocratic Status Quo | 50 | 48 | 15 | 42 | 45 | 52 | 37 | 63 | 51 | 45% | 46% | 12.3 |
| The Fracture | 6 | 6 | 5 | 8 | 10 | 8 | 13 | 8 | 10 | 8% | 8% | 2.5 |
| Q3: China: Economic & Political Model | ||||||||||||
| Pragmatic Re-Opening | 15 | 13 | 10 | 22 | 15 | 8 | 22 | 20 | 15 | 16% | 16% | 4.7 |
| Succession Crisis | 7 | 3 | 5 | 8 | 5 | 12 | 12 | 18 | 5 | 8% | 8% | 4.5 |
| Fortress China | 32 | 26 | 35 | 25 | 30 | 30 | 28 | 37 | 40 | 31% | 32% | 5.0 |
| Managed Continuity | 46 | 58 | 50 | 45 | 50 | 50 | 38 | 25 | 40 | 45% | 45% | 8.9 |
| Q4: Physical Climate Manifestation | ||||||||||||
| Crisis-Driven Adaptation | 22 | 22 | 35 | 22 | 18 | 24 | 24 | 9 | 11 | 21% | 21% | 7.5 |
| Current Trajectory | 58 | 72 | 55 | 48 | 59 | 47 | 49 | 68 | 63 | 58% | 58% | 8.8 |
| Resilience Victory | 13 | 1 | 5 | 12 | 13 | 18 | 13 | 3 | 16 | 10% | 10% | 5.7 |
| Benign Plateau | 7 | 5 | 5 | 18 | 10 | 11 | 15 | 20 | 10 | 11% | 11% | 5.2 |
Hover over outcome names for descriptions. Purple = Thinking AI models, Blue = Average of 6 legacy AI models, Orange = Human forecasters (Avg Hum Ex-Sam = Rimas, Adam, Ty, Sol, Chris). Green = highest, Red = lowest. Wtd Avg uses model-specific weights.
Figure 5 Forecaster Responses (Q5–Q8)
| Outcome | GPT 5 Pro | Sonnet 4 | Gemini 2.5 | Opus 4 | Deep Seek | Grok 4 |
Avg Legacy |
Sam | Avg Hum Ex-Sam |
Simple Avg |
Wtd Avg |
σ |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Q5: Global Energy Trajectory | ||||||||||||
| Security First | 27 | 20 | 20 | 25 | 30 | 42 | 31 | 38 | 35 | 30% | 30% | 7.2 |
| Green Tech Acceleration | 48 | 65 | 45 | 50 | 35 | 15 | 27 | 20 | 25 | 37% | 38% | 15.7 |
| Scarcity Scramble | 14 | 10 | 30 | 15 | 15 | 13 | 21 | 40 | 24 | 20% | 20% | 9.2 |
| Nuclear Renaissance | 11 | 5 | 5 | 10 | 20 | 30 | 21 | 2 | 16 | 13% | 13% | 8.8 |
| Q6: India: The Modi Legacy | ||||||||||||
| Hindu Nationalist Peak | 26 | 15 | 10 | 36 | 40 | 31 | 35 | 30 | 30 | 28% | 28% | 9.5 |
| Federal Fissure | 18 | 25 | 15 | 35 | 25 | 15 | 24 | 30 | 25 | 24% | 23% | 6.5 |
| Economic Pivot | 34 | 45 | 45 | 19 | 20 | 36 | 24 | 25 | 27 | 31% | 31% | 9.8 |
| Opposition Resurgence | 22 | 15 | 30 | 10 | 15 | 18 | 17 | 15 | 18 | 18% | 18% | 5.4 |
| Q7: Japan: Strategic Recalibration | ||||||||||||
| Rearmed Ally | 46 | 62 | 60 | 38 | 45 | 32 | 32 | 30 | 18 | 40% | 41% | 14.3 |
| Tech-First Strategy | 32 | 23 | 20 | 12 | 25 | 18 | 23 | 20 | 15 | 21% | 22% | 5.5 |
| Managed Decline | 12 | 10 | 5 | 19 | 10 | 12 | 18 | 20 | 41 | 16% | 16% | 10.6 |
| Balanced Hedger | 10 | 5 | 15 | 31 | 20 | 38 | 28 | 30 | 26 | 23% | 21% | 10.6 |
| Q8: AI & Technology Revolution | ||||||||||||
| Productivity Boom | 42 | 23 | 25 | 41 | 45 | 14 | 34 | 15 | 17 | 28% | 30% | 12.2 |
| Winner-Takes-All | 26 | 55 | 50 | 24 | 25 | 56 | 32 | 50 | 21 | 38% | 37% | 14.5 |
| Regulated Utility | 19 | 5 | 10 | 15 | 20 | 11 | 19 | 6 | 16 | 13% | 14% | 5.2 |
| Progress Plateau | 13 | 17 | 15 | 20 | 10 | 19 | 16 | 29 | 46 | 21% | 20% | 11.3 |
Purple = Thinking AI models, Blue = Average of 6 legacy AI models, Orange = Human forecasters. Notable: Human forecasters (Avg Hum Ex-Sam) are much more skeptical of AI progress—46% on Progress Plateau vs AI's ~16%. Japan's Managed Decline shows 41% human consensus vs 11% AI average. Wtd Avg uses model-specific weights.
Figure 6 Forecaster Responses (Q9–Q13)
| Outcome | GPT 5 Pro | Sonnet 4 | Gemini 2.5 | Opus 4 | Deep Seek | Grok 4 |
Avg Legacy |
Sam | Avg Hum Ex-Sam* |
Simple Avg |
Wtd Avg |
σ |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Q9: Semiconductor Supply Chains | ||||||||||||
| Taiwan Hyper-Concentration | 24 | 38 | 50 | 55 | 20 | 48 | — | 5 | 36 | 35% | 33% | 16.8 |
| Successful Diversification | 51 | 27 | 30 | 15 | 45 | 35 | — | 70 | 33 | 38% | 41% | 16.4 |
| Crisis-Driven Distribution | 14 | 20 | 15 | 20 | 20 | 9 | — | 15 | 19 | 17% | 16% | 3.9 |
| Crisis-Driven Slowdown | 11 | 15 | 5 | 10 | 15 | 8 | — | 10 | 12 | 11% | 11% | 3.2 |
| Q10: Critical Materials | ||||||||||||
| Chinese Chokepoint | 17 | 40 | 20 | 24 | 20 | 25 | — | 8 | 25 | 22% | 21% | 8.8 |
| Structural Deficit | 43 | 22 | 50 | 35 | 35 | 30 | — | 24 | 36 | 34% | 35% | 9.2 |
| Balanced Expansion | 32 | 23 | 25 | 28 | 25 | 32 | — | 50 | 27 | 30% | 31% | 8.0 |
| Demand Disappointment | 8 | 15 | 5 | 13 | 20 | 13 | — | 18 | 12 | 13% | 13% | 4.7 |
| Q11: US Dollar Global Role | ||||||||||||
| Permanent Dollar | 45 | 47 | 35 | 32 | 38 | 52 | — | 12 | 37 | 37% | 38% | 12.0 |
| Dollar Downturn | 17 | 16 | 15 | 20 | 22 | 12 | — | 40 | 21 | 20% | 20% | 8.5 |
| Fragmented System | 28 | 19 | 40 | 33 | 28 | 24 | — | 20 | 27 | 27% | 28% | 6.6 |
| Crisis Flight to Safety | 10 | 18 | 10 | 15 | 12 | 12 | — | 28 | 15 | 15% | 14% | 5.6 |
| Q12: Interest Rates & Cost of Borrowing | ||||||||||||
| High Rate Regime | 21 | 28 | 15 | 28 | 34 | 28 | — | 5 | 24 | 23% | 23% | 8.6 |
| Volatile Refinancing Trap | 27 | 18 | 40 | 22 | 38 | 18 | — | 30 | 29 | 28% | 28% | 7.9 |
| New Normal Plateau | 44 | 47 | 35 | 39 | 28 | 42 | — | 60 | 37 | 42% | 42% | 9.0 |
| Return to Zero | 8 | 7 | 10 | 11 | 0 | 12 | — | 5 | 10 | 8% | 8% | 3.7 |
| Q13: Global Inflation & Real Cost of Capital | ||||||||||||
| Synchronized Soft Landing | 47 | 47 | 35 | 26 | 45 | 52 | — | 14 | 31 | 37% | 39% | 12.7 |
| Stagflation | 17 | 17 | 25 | 20 | 15 | 9 | — | 15 | 23 | 18% | 17% | 5.0 |
| Persistent Inflation | 28 | 27 | 30 | 40 | 30 | 27 | — | 58 | 36 | 35% | 33% | 9.8 |
| Deflationary Undershoot | 8 | 9 | 10 | 14 | 10 | 12 | — | 13 | 10 | 11% | 11% | 1.9 |
Purple = Thinking AI models, Orange = Human forecasters. Legacy AI models not surveyed on Q9-Q13 (shown as —). *Avg Hum Ex-Sam extrapolated from Q1-Q8 response patterns. Sam's distinctive views: 70% on Successful Diversification, only 12% on Permanent Dollar, 58% on Persistent Inflation, 5% on High Rate Regime. Wtd Avg uses model-specific weights.
The Complete Picture
With scenarios defined, securities chosen, and probabilities assigned, we can now construct the complete return distribution for each security across all 52 scenario states. This isn't a single expected return—it's a full probability-weighted distribution showing not just the mean but the shape of possible outcomes.
Figure 7 Scenario Return Distributions (52 Scenarios × 12 Securities)
6-year expected returns for 12 representative securities across 52 scenario states. Rows colored by type: crisis, stress, positive, breakthrough. Toggle to see volatility and skewness.
The pattern is striking: crisis scenarios systematically hurt risk assets while benefiting safe havens. Long Treasuries (TLT) show +20% to +25% expected returns in Constitutional Crisis and Climate Catastrophe scenarios—exactly when equity portfolios need protection. Gold (GLD) provides even more dramatic crisis performance, with +30% to +40% expected returns in the worst scenarios. Meanwhile, the AI Breakthrough scenario shows +60% for US Tech and +50% for Taiwan—the asymmetric upside that justifies technology exposure despite volatility.
But expected returns are only part of the picture. The volatility view reveals which scenarios create the most uncertainty within the scenario itself. Crisis scenarios show elevated volatility across the board—Constitutional Crisis pushes FTEC volatility to 32% and FXI to 38%. The skewness view is perhaps most interesting: it shows the direction of tail risk. In crisis scenarios, equities show negative skew (fat left tails—crashes more likely than spikes), while gold shows positive skew (fat right tails—potential for dramatic upside).
Figure 8 Security Correlation Matrix (12 Representative Securities)
Pairwise correlations across 52 scenario states for 12 representative securities spanning US sectors, international equities, fixed income, and alternatives. Hover to see full distribution. Key diversifiers: TLT shows -0.41 to -0.55 correlation with equities; Gold shows -0.46 with Tech.
The correlation structure reveals the portfolio's diversification architecture. US equity sectors form a tight cluster with correlations above 0.7—rotating among them provides limited diversification. Long Treasuries (TLT) show the most dramatic negative correlations with equities: -0.35 to -0.45 with growth and cyclical sectors. This is the structural diversification that makes a 60/40 portfolio work in normal times and crisis scenarios alike.
Gold occupies a unique position: near-zero or slightly negative correlation with most equities, positive correlation with Treasuries (0.35) and TIPS (0.45), and positive correlation with commodities (0.40). It's not just an inflation hedge—it's genuinely uncorrelated with the dominant equity risk factors. Taiwan (EWT) correlates 0.82 with US Tech but only 0.55 with India and 0.60 with China—it's semiconductor exposure, not generic EM exposure.
What This Enables
This framework isn't just a different way to get the same answer. It enables analyses that backward-looking approaches can't perform.
Stress testing with coherence. Traditional stress tests shock individual variables: "what if rates rise 200 bps?" But rates don't rise in isolation. A 200 bps rate shock probably comes with higher inflation, changed Fed policy, and second-order effects throughout the economy. Scenario-based stress testing shocks coherent worlds, not isolated variables. The results are more realistic.
Explicit belief attribution. When a portfolio underperforms, you can trace it to specific beliefs that didn't pan out. "I underweighted defense because I assigned only 10% to Cold War scenarios; the actual probability was higher." This is infinitely more useful than "the factors underperformed"—it tells you what to update.
Robust optimization. Instead of optimizing for a single expected return, you can optimize for performance across scenarios. Maximin strategies maximize minimum performance. Minimax regret strategies minimize worst-case regret. These approaches sacrifice some expected return for resilience to belief errors.
Dynamic rebalancing triggers. As events unfold, scenario probabilities shift. You can define rebalancing triggers: "if Cold War probability exceeds 25%, increase defense allocation." This turns portfolio management into Bayesian updating rather than calendar-based rebalancing.
Cross-asset class allocation. The framework doesn't just rotate within equities—it shifts between equities, fixed income, and alternatives based on scenario probabilities. A rising probability of Stagflation shifts allocation toward TIP and GLD and away from TLT and growth equities. A rising probability of Tech Dominance shifts toward FTEC and EWT and away from defensive dividend stocks. The 30-security universe enables genuine macro positioning, not just sector tilts.
The framework is complete. What remains is implementation.