read
articles
VIP Basil Halperin

VIP Basil Halperin

📝 ARTICLE INFORMATION

  • Title: VIP Basil Halperin
  • Person: Basil Halperin
  • Position: Assistant Professor of Economics, University of Virginia (2025–); formerly Postdoctoral Fellow, Stanford Digital Economy Lab (2024–2025)
  • Platform: basilhalperin.com (research, essays, blog)
  • Date: October 16, 2025
  • Primary Focus: Monetary economics, macroeconomic growth, AI economics
  • Word Count: Approximately 3,700 words

🎯 HOOK

A 27-year-old MIT-trained economist just proved that financial markets are telling us something crucial about AI timelines: either transformative AI won’t arrive for 30+ years, or $1 trillion is lying on the table waiting to be arbitraged.

💡 ONE-SENTENCE TAKEAWAY

Basil Halperin uses real interest rates as an information aggregator to argue that efficient markets decisively reject short AI timelines; if they’re wrong, both philanthropists and traders have enormous arbitrage opportunities to exploit, which means the disagreement is falsifiable and has real stakes.

📖 SUMMARY

Basil Halperin is a macroeconomist building bridges between core economic theory, financial markets, and transformative AI. His work occupies a distinctive position: rigorous academic research grounded in mathematical models, but translated into practical implications for policy, philanthropy, and financial decision-making.

His PhD thesis at MIT (2024), supervised by Angeletos and Werning, comprises three essays on monetary policy and economic growth. The first essay, joint with Daniele Caratelli, challenges the standard New Keynesian prescription for optimal monetary policy. Textbook economics says central banks should keep inflation stable after supply shocks; the inflation targeting regime. But under menu costs (the realistic observation that firms pay to change prices), the optimal policy is different: inflation and output should move inversely after sectoral shocks. After a negative productivity shock, inflation should rise, and the central bank should target nominal wages rather than inflation. This produces a 0.32% welfare improvement over standard policy, which is substantial when aggregated across an economy.

The second essay reexamines monetary policy at the zero lower bound (ZLB), challenging the conventional wisdom that monetary policy becomes impotent. Halperin argues that if ZLB episodes are repeated games (empirically realistic) rather than one-shot events, reputational effects solve the time consistency problem that typically constrains policy. He also shows that fiscal policy faces the same time consistency problem at the ZLB, suggesting fiscal policy power has been overrated. Austerity is stimulus in representative agent models once you account for consumption smoothing.

The third essay is co-authored with Trevor Chow and J. Zachary Mazlish. It applies economic theory to the AI timeline debate. This work has generated the most public engagement.

The AGI and EMH Argument:

Halperin et al. observe that under standard economic models, transformative AI would produce very high real interest rates, whether aligned or unaligned. Aligned AI accelerating growth at 30%+ annually would make people want to save less and consume today (since they’ll be rich tomorrow), pushing rates up. Unaligned AI creating extinction risk would also push rates up (people save less if they might not have a future to save for). Yet 30-year real interest rates are currently low.

From this they derive two possibilities: either (1) markets are efficient information aggregators, and low rates reflect that short AI timelines are unlikely, or (2) markets are radically inefficient, meaning there’s $1+ trillion in arbitrage opportunities.

The work is sophisticated in multiple ways. It provides empirical evidence that growth expectations actually correlate with higher real interest rates across countries and historical periods. It shows that simply looking at stock prices is misleading (public companies might capture only part of AI value; growth could depress equity valuations). Real interest rates (the price of borrowing across time) are the better information aggregator.

The essay engages with critiques seriously. Halperin acknowledges that real interest rates are murky (citing Tyler Cowen’s Third Law: “all propositions about real interest rates are wrong”). He notes that policy tightening, inflation, and geopolitical risk could explain recent rate increases independent of AI expectations. He even created Metaculus forecasting questions to test the framework.

Crucially, he recognizes the implications: “If short timelines are your true belief in your heart of hearts, and not merely a belief in a belief, then you should seriously consider how much money you could earn here and what you’ll do with potential profits.” This transforms the disagreement from academic debate into testable, financial terms.

Beyond the Flagship Essay:

Halperin’s other essays show range. “It was a mistake to switch to sticky price models from sticky wage models” argues that 1970s-era shifts in economic modeling caused unnecessary confusion and wrong policy conclusions. “Newcomb’s problem is just a standard time consistency problem” translates philosophical problems into monetary policy language, showing that standard decision theory solves what philosophers treat as exotic. “Recessions are always and everywhere caused by monetary policy” defends a strong claim about monetary transmission mechanisms.

The essays share stylistic traits: they combine mathematical rigor with prose clarity. Halperin explains concepts (“menu costs,” “nominal wage targeting”) without oversimplifying. He engages counterarguments. He cites evidence. He acknowledges limitations. He avoids jargon when plain speech works. Most importantly, he makes implications concrete. Theoretical insights connect to real policy debates, real asset prices, real philanthropic strategy.

The Research Program:

His current work at Stanford Digital Economy Lab and soon at University of Virginia focuses on AI and macroeconomic growth. He’s directing the Stripe Economics of AI Fellowship, curating research on how AI impacts productivity, distribution, labor markets, and growth. This represents broader institutional recognition: major economic funders are betting that his framework (connecting core macro theory to AI economic implications) is productive.

The through-line in his work: economic mechanisms matter more than intuition. The intuition says “monetary policy can’t work at zero rates” but the math shows reputational effects overcome the constraint. The intuition says “growth expectations raise stock prices” but market prices (real rates) show otherwise. The intuition says “AI timelines are pure speculation” but financial data provides leverage.

His career trajectory, Uber (applying theory), AQR (testing on real markets), MIT PhD (formalizing insights), Stanford (translating to broader impact), academic position (institutionalizing research), reflects someone translating between theory and practice without losing rigor at either end.

🔍 INSIGHTS

Core Insights

  • Real Interest Rates as Information Aggregator: Asset prices reflect all public information efficiently. Real interest rates, being foundational to all asset pricing, reveal what markets believe about long-term growth and extinction risk. Low real rates imply low probability of near-term transformative AI.
  • Monetary Policy Untapped: Standard monetary policy tools work better at the zero lower bound than textbook wisdom suggests when you account for repeated-game dynamics and reputation. Central banks have been undershooting optimal policy.
  • Menu Costs Change Optimal Policy: When firms pay to change prices, optimal policy is not inflation targeting but nominal wage targeting. This explains historical regime shifts and suggests policy recommendations differ from textbook prescriptions.
  • Growth and Interest Rates Correlate Positively: Contrary to some recent claims, higher expected growth causes higher long-term real interest rates due to consumption smoothing. This empirical relationship holds across countries and time periods.
  • AI Timeline Disagreements are Falsifiable: The framework makes AI timeline beliefs testable through financial markets. If you truly believe short timelines, you have arbitrage opportunities. This transforms disagreement from unfalsifiable speculation to testable prediction.
  • Philosophical Problems Map to Economics: Newcomb’s problem, causal decision theory, and other philosophical frameworks have direct parallels in economic models. Economic language often clarifies what philosophical language obscures.
  • Welfare-Improving Policy Exists: Small improvements in policy design (0.32% welfare gain) compound massively across populations and time. Monetary policy research has practical, non-trivial returns.

How This Connects to Broader Trends/Topics

  • AI Economics: Bridges macroeconomic theory to AI forecasting, connecting growth theory to tech timelines in a way that generates testable predictions.
  • Monetary Policy Reform: Contributes to growing critique of inflation targeting and exploration of alternatives (nominal GDP targeting, nominal wage targeting).
  • Effective Altruism: Provides framework for philanthropists to evaluate AI timeline belief-consistency through financial decision-making.
  • Market Efficiency: Tests EMH rigorously in context of novel, hard-to-forecast events, showing where markets might aggregate information well or poorly.
  • Academic-Industry Bridge: Career path demonstrates how academic rigor combines with real-world trading, tech work, and policy influence.

🛠️ FRAMEWORKS & MODELS

Real Interest Rates as AI Timeline Indicator

  • Explanation: Standard asset pricing theory says real interest rates reflect expectations of future growth and consumption. Transformative AI (aligned or unaligned) would massively increase either growth expectations or extinction risk, both of which increase real rates.
  • Mechanism:
    • Aligned rapid AI: Future consumption high → save less today → rates rise
    • Unaligned AI risk: Future might not exist → save less today → rates rise
    • Either scenario produces high real rates
  • Empirical Test: 30-year real rates are low, suggesting markets don’t expect near-term transformative AI
  • Implications:
    • EMH interpretation: Trust markets, assume long timelines
    • Market inefficiency interpretation: $1T+ arbitrage opportunity exists
  • Application: Track real interest rates as live forecast of AI timeline expectations. If rates rise sharply without inflation/policy explanation, market expectations of near-term AI may have updated.
  • Significance: Transforms AI timeline from unfalsifiable belief to testable financial prediction with monetary stakes.

Optimal Monetary Policy Under Menu Costs

  • Explanation: When firms pay to change prices (menu costs), standard inflation targeting is suboptimal. Instead, optimal policy targets nominal wages.
  • Logic:
    • Inflation targeting forces firms to adjust prices after sectoral shocks, incurring deadweight menu costs
    • Nominal wage targeting allows relative prices to adjust while firms pay menu costs only when necessary
    • Result: fewer firms adjust → lower total menu costs → higher welfare
  • Quantitative Result: 0.32% welfare improvement in baseline calibration
  • Policy Implications: Central banks should shift from inflation targeting to nominal wage targeting
  • Application: Evaluates whether current policy regimes are theoretically optimal; suggests reform opportunities
  • Significance: Small theoretical improvements compound massively at economy scale, justifying research on monetary policy design.

Time Consistency in Repeated Games vs. One-Shot Games

  • Explanation: Monetary policy at zero lower bound traditionally faces time consistency problem: central bank incentive to deviate from policy promise once private expectations set.
  • Refinement: If ZLB episodes are repeated games (empirically true), reputation effects solve time consistency. Deviating once costs future credibility.
  • Application: Central bank can commit credibly to policy through implicit reputation mechanism. Problem is not insoluble.
  • Significance: Expands policy option space at ZLB; suggests monetary policy less constrained than standard models suggest.

Market Efficiency Decomposition

  • Explanation: Test whether markets efficiently aggregate information by examining price signals that should reveal AI timeline expectations.
  • Analysis:
    • Stock prices: Ambiguous (could reflect growth, profit distribution, or risk aversion)
    • Real interest rates: Clear signal (all assets discounted using these rates; hard to arbitrage away)
  • Advantage of real rates: No-arbitrage condition forces these to reflect true market beliefs more clearly than equity prices
  • Application: When predicting technology timelines, use financial prices that have clearest theoretical implications and fewest confounds.
  • Significance: Demonstrates how to extract testable predictions from financial markets for novel events.

Newcomb’s Problem as Time Consistency Problem

  • Explanation: Philosophical puzzle about causal vs. evidential decision theory maps directly to economic time consistency problem.
  • Parallel:
    • Philosophical: Should you decide based on what decision-maker predicts you’ll do (evidential) or on causal consequences of your choice (causal)?
    • Economic: Should central bank decide based on state people expect given announcement (evidential/reputation) or on actual consequences of money supply (causal)?
  • Significance: Shows economic language often clarifies philosophical problems; standard economic decision theory solves “exotic” philosophical puzzles.
  • Application: Translate philosophical disagreements to economic language to test empirically.

💬 QUOTES

  • “Short AI timelines would cause real interest rates to be high. However, 30-year real interest rates are low.”

    • Context: Opening of AGI and EMH essay; statement of the core puzzle driving the entire argument.
    • Significance: Encapsulates the core insight: either markets know something we don’t about AI timelines, or there’s massive arbitrage opportunity.

  • “If real interest rates are wrong, all financial assets are mispriced. If real interest rates ‘should’ rise three percentage points or more, that is easily hundreds of billions of dollars worth of revaluations.”

    • Context: Emphasizing the scale of potential market inefficiency.
    • Significance: Makes the disagreement concrete and testable. If you believe markets are wrong, you can profit.

  • “If short timelines are your true belief in your heart of hearts, and not merely a belief in a belief, then you should seriously consider how much money you could earn here and what you’ll do with potential profits.”

    • Context: Challenging readers to demonstrate consistency between AI timeline beliefs and financial behavior.
    • Significance: Shifts debate from academic to practical; reveals which beliefs people genuinely hold vs. performatively express.

  • “Markets are decisively rejecting the shortest possible timelines of 0-10 years.”

    • Context: Quantitative claim about market interpretation of real rate data.
    • Significance: Makes timeline prediction specific and falsifiable.

  • “We view our argument as the best existing outside view evidence on AI timelines – but also as only one model among a mixture of models that you should consider when thinking about AI timelines.”

    • Context: Epistemically humble conclusion to major claim.
    • Significance: Demonstrates good faith engagement with uncertainty; avoids overconfidence.

  • “All propositions about real interest rates are wrong.”

    • Context: Citing Tyler Cowen’s Third Law as caveat to entire framework.
    • Significance: Self-aware acknowledgment that core concept is theoretically murky; doesn’t invalidate framework but tempers confidence.

⚡ APPLICATIONS

For AI Researchers & Forecasters

  • Use Real Interest Rates as Live Signal: Monitor 30-year real interest rates (inflation-linked bond yields) as market forecast of AI timelines. Significant increase suggests market updated on near-term AGI probability.
  • Track Growth Expectations Separately: Use survey data on inflation expectations and nominal GDP forecasts to isolate AI-specific belief updates from other factors.
  • Design Testable Predictions: Follow Halperin’s model: make specific predictions about financial prices if your AI timeline belief is correct, then track whether prices move as predicted.
  • Exploit Inconsistencies: If you believe short timelines but markets don’t, structure financial trades to capture this belief. Real skin-in-game reveals true conviction.

For Central Banks & Policymakers

  • Reconsider Inflation Targeting: Examine menu cost evidence and wage-targeting literature. Determine whether switching regimes would improve welfare. Test on pilot basis if feasible.
  • Exploit Zero Lower Bound Less Passively: Reputation effects solve time consistency problems more effectively than standardly modeled. Design credible commitment mechanisms (forward guidance, price-level targets) that leverage reputational stake.
  • Monitor AI Expectations Through Rates: Real interest rates contain information about long-term growth expectations. If rates rising sharply, extract signal about what markets expect for AI-driven growth.

For Philanthropists & AI Safety Organizations

  • Align Financial Strategy with Timeline Beliefs: If you believe short timelines, Halperin’s framework gives explicit permission (and financial logic) to borrow cheaply today, spend on safety now, rather than saving for indefinite future.
  • Test Consistency: Examine your timeline beliefs by asking: am I willing to borrow at current real rates bet on my timeline forecast? If not, your stated timeline belief is weaker than you think.
  • Use as Reference Class: Even if you disagree with EMH, use market prices as one reference class. Desert extreme timeline confidence in absence of countervailing evidence stronger than financial markets provide.

For Investors & Asset Allocators

  • Real Rate Sensitivity Analysis: Model portfolio under different real-rate scenarios. If AI timelines shorten unexpectedly, real rates spike, affecting all asset valuations simultaneously.
  • AI-Conditional Assets: Evaluate whether current holdings appropriately hedge AI scenario. If you believe short timelines, position portfolio to benefit from rate spike.
  • Monitor Leading Indicators: Track indicators Halperin identifies (growth expectations via surveys, nominal rates across countries, corporate investment decisions) as early warning of market updating on AI timelines.

For Economists & Researchers

  • Test Menu Cost Theory: Empirically validate nominal wage targeting vs. inflation targeting across monetary policy regimes. NGDP targeting debate could be resolved through careful institutional comparison.
  • Improve Real Rate Theory: Halperin acknowledges real rates are theoretically murky. Research that clarifies determinants of real rates (demographics, growth expectations, risk preferences) would improve forecasting power.
  • Extend to Other Tech Timelines: Apply Halperin’s real-interest-rate framework to other transformative technologies (biotech, fusion, quantum). See whether financial markets provide usable signals for tech timeline forecasting generally.

📚 REFERENCES

PhD Thesis & Core Research

  • Halperin, B. (2024). “Essays in Monetary Policy and Growth.” PhD thesis, MIT. Advisors: George-Marios Angeletos, Iván Werning.
    • Chapter 1: “Optimal Monetary Policy Under Menu Costs” (with Daniele Caratelli)
    • Chapter 2: “Reexamining Optimal Policy in the New Keynesian Liquidity Trap”
    • Chapter 3: “Transformative AI, Existential Risk, and Real Interest Rates” (with Trevor Chow, J. Zachary Mazlish)

Published & Accepted Papers

  • Halperin, B., Ho, B., List, J.A., & Muir, I. (2022). “Toward an understanding of the economics of apologies: Evidence from a large-scale natural field experiment.” The Economic Journal, 132(641), 273–298.

    • Analyzed 1.5M Uber ridesharing apologies; found money speaks louder than words for service recovery

  • Halperin, B. & Mazlish, J.Z. “Overreaction and forecast horizon”

    • Established that expectations overreact at 2+ year horizons across 89 countries, 35 years of data

Essays & Public Work

  • “AGI and the EMH: markets are not expecting aligned or unaligned AI in the next 30 years” (2023)

    • Featured in: The Economist, Vox Future Perfect, Marginal Revolution
    • Received first prize in Open Philanthropy AI Worldviews Contest

  • “Against using stock prices to forecast AI timelines” (2023)

    • Appendix to AGI and EMH essay
    • Argues real interest rates provide better signal than equity prices

  • Essays on monetary policy:

    • “It was a mistake to switch to sticky price models from sticky wage models”
    • “Newcomb’s problem is just a standard time consistency problem”
    • “Recessions are always and everywhere caused by monetary policy”
    • “Explaining Tyler Cowen’s Third Law”
    • “The ‘Efficient Restaurant Hypothesis’: a mental model for finance (and food)”
    • “Yes, markets are efficient – and yes, stock prices are predictable”

Positions & Affiliations

  • Assistant Professor of Economics, University of Virginia (2025–)
  • Associate Director, Economics of Transformative AI Initiative, UVA (2025–)
  • Director, Stripe Economics of AI Fellowship (2025–)
  • Digital Fellow, Stanford Digital Economy Lab (2024–2025)
  • Postdoctoral Fellow, Stanford Digital Economy Lab (2024–2025)
  • Research Economist, Uber/Ubernomics (2017–2018)
  • Quantitative Research Analyst, AQR Capital Management (2015–2016)

Academic Background

  • B.S. Mathematics, Economics, Chinese, University of Chicago (2015)
  • PhD Economics, MIT (2024)

Referenced Works & Frameworks Engaged

  • Angeletos, G.M., & Werning, I. (advisors): Core monetary policy and macro theory
  • Cowen, T.: “Tyler’s Third Law” on real interest rates; general macro commentary
  • Caratelli, D.: Menu costs and optimal monetary policy collaboration
  • Chow, T. & Mazlish, J.Z.: AI economics and asset pricing collaboration
  • New Keynesian models (Calvo pricing, sticky wages, representative agent models)
  • Asset pricing theory (consumption smoothing, real rates, discount factors)
  • Efficient markets hypothesis and financial economics

Resources for Further Engagement

  • Website: basilhalperin.com (research, essays, contact info)
  • Twitter: @basilhalperin
  • Metaculus forecasting questions on monetary policy
  • Stripe Economics of AI Fellowship application process

⚠️ QUALITY & TRUSTWORTHINESS NOTES

Strengths

  • Rigorous Formalization: Mathematical models are explicitly stated, assumptions transparent, allowing verification and critique.
  • Empirical Grounding: Claims backed by data (89-country survey analysis, historical interest rate patterns, cross-country evidence).
  • Real-World Testing: Tested theories at Uber (behavioral experiments at scale) and AQR (quantitative trading), not just theoretically.
  • Epistemic Humility: Acknowledges Tyler Cowen’s Third Law applies to own work; notes multiple confounds for real rate interpretation; treats framework as “one model among many.”
  • Engagement with Critique: Responded to LessWrong/EA Forum objections seriously; created Metaculus forecasting questions to test framework; published appendix addressing limitations.
  • Transparent About Stakes: Makes assumptions clear (EMH as baseline, specific preference parameters in models) and shows how results change under alternative assumptions.

Considerations

  • Young Researcher: Just completed PhD (2024); limited track record of policy influence or paper citations in academic literature (may increase with time).
  • Efficient Markets Hypothesis: Framework depends on EMH as reasonable prior. Growing evidence of market inefficiencies in specific domains could weaken central argument.
  • AI Timeline Specificity: Framework works best for 30+ year timelines where financial markets have forward-looking data. Less useful for very-near-term predictions (5-10 years).
  • Real Rate Murk: Halperin himself acknowledges theoretical uncertainty about real rate determinants. Multiple factors (demographics, risk, policy) confound interpretation.
  • Limited Alternative Tests: While framework makes predictions, limited independent verification so far (waiting for actual AI developments or market movements).
  • Assumption Sensitivity: Welfare results (0.32% improvement) depend on specific parameter choices. Different menu cost calibrations could change conclusions.

Accuracy Assessment

  • Economic Theory: Descriptions of New Keynesian models, menu cost theory, time consistency problems align with mainstream macroeconomics.
  • Empirical Claims: Cross-country growth/interest rate correlations appear accurately reported; survey data methodology sound.
  • AI Mechanism: Logic that either AI scenario (aligned rapid growth or extinction risk) increases real rates is standard asset pricing result, not novel but applied cleverly.
  • Replication Likelihood: Framework is testable; actual replication likelihood depends on follow-up research.

Trust Indicators

  • Institutional Recognition: MIT PhD supervision, Stanford postdoc, now-tenured at UVA, directing Stripe fellowship.Major institutions are betting on his framework.
  • External Validation: Open Philanthropy AI Worldviews Contest first prize; featured in The Economist, Vox; cited by leading researchers.
  • Intellectual Community: Engages seriously with LessWrong, EA Forum, academic collaborators; responds to critiques.
  • Professional Trajectory: Credible path from trading/tech to academic research to policy influence suggests genuine insights rather than hype.

Potential Limitations

  • EMH Assumption: If markets are more irrational than assumed, entire framework loses force. COVID experience showed markets can be surprisingly slow to process novel information.
  • Confounding Variables: Recent real rate rises could reflect policy tightening, demographic shifts, or risk premium increases rather than AI beliefs.
  • Selection of Signal: Why real rates rather than other long-duration assets? Framework doesn’t fully justify this choice against alternatives.

Find this work at: basilhalperin.com | @BasilHalperin on X | Department of Economics, University of Virginia

Crepi il lupo! 🐺