AI may reduce average inefficiency but increase tail inefficiency

Overview

The report centers on whether AI makes markets more or less efficient. The core conclusion is that AI is likely to improve market microstructure efficiency, including faster information processing, better execution, broader data coverage, and reduced behavioral biases; however, it may also intensify macro and market-structure inefficiency, such as crowded trading, model homogeneity, rising concentration, synthetic information shocks, and more severe narrative-driven reversals. Therefore, AI may produce lower average inefficiency but higher tail inefficiency.

Core views

In normal market conditions, AI may improve liquidity, shorten reaction times, reduce obvious mispricing, and compress simple alpha; but in bubble or stress environments, similar models are likely to identify the same winners and losers, leading to higher correlation, turnover, speed of deleveraging, and liquidity gaps. The report believes competitive advantage will shift from arbitraging public signals toward proprietary data, workflow integration, differentiated judgment, and stronger execution.

Analysis framework

The report uses a bidirectional evidence framework: on one side it lists evidence that AI improves market efficiency, including reductions in forecast disagreement, price discovery, research coverage, and behavioral bias cases; on the other side it analyzes AI-induced inefficiency through crowding effects, systemic risk, synthetic information shocks, potential collusive behavior, bid-ask spread shifts, and reflexivity issues. Coverage spans U.S. equities, Asian markets, EM small caps, India mid caps, Taiwan, Korea, Japan, and yen carry/arbitrage cases.

Methodology notes

• Market microstructureInformation asymmetry and price discovery

  When information processing costs fall and the speed at which information is incorporated into prices rises, market microstructure efficiency improves.

  The report uses declines in EPS forecast dispersion, convergence in earnings surprises, and faster price reactions after Fed minutes as evidence of AI improving information efficiency.

• Behavioral financeBehavioral biases and AI agents

  AI agents may reduce overconfidence, anchoring, loss aversion, and herding.

  The report cites experimental studies showing AI agents making rational decisions at 61% to 97%, higher than human participants at 46% to 51%, but also notes that under profit-maximization incentives, AI agents may still optimally follow the crowd.

• Market structure riskCrowded trading and model homogeneity

  When many investors use similar models, data, and objectives, convergent signals can amplify one-way positioning and liquidity pressure.

  The report points to the August 2024 Nikkei 225 selloff, VIX above 60, and yen carry unwinds as examples of machine-driven strategies mechanically de-risking under stress.

• ReflexivitySoros reflexivity

  Model outputs affect trades, trades change prices and expectations, which then reinforce future model signals.

  The report argues that AI is both an investment tool and a narrative driver that may push concentration, momentum, and valuation extremes in Asian tech stocks.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• Large-cap, highly covered stocks

  AI improves information processing and forecast convergence speed

  Strengths

  Information asymmetry declines, and EPS forecast dispersion and earnings surprises may narrow.

  Weaknesses

  The alpha from simple, transparent, widely observable public signals decays more quickly.

  Comparison

  Compared with small and mid caps, information efficiency is already relatively high, and AI's incremental benefit may be seen more in speed and execution.

  Risks

  Reversals may be sharper when consensus becomes overly uniform.

• EM small caps and India mid caps

  AI lowers marginal coverage cost for research

  Strengths

  Broader coverage, improved liquidity, and faster discovery of prior neglect premia.

  Weaknesses

  Opportunities from low prior coverage may be arbitraged away more quickly.

  Comparison

  Compared with large caps, AI's marginal impact on coverage expansion is more pronounced.

  Risks

  Data quality, liquidity, and model misclassification may still limit efficiency gains.

• Taiwan and Korea tech/AI-agent sectors

  AI is both a tool and the dominant market narrative

  Strengths

  Strong momentum supported by earnings expectations and thematic flows.

  Weaknesses

  Market-cap concentration, valuations, crowding, and analyst optimism are at extreme levels.

  Comparison

  The report says Taiwan tech accounts for about 85% of market cap and Korea tech about 70%, levels of concentration rare in Asian markets.

  Risks

  Momentum reversals, rising correlations, valuation drawdowns, and narrative fade.

• Yen carry and leveraged trading

  Systematic strategies and crowded positioning can amplify shocks

  Strengths

  Provides carry and liquidity in stable environments.

  Weaknesses

  In stress periods, mechanical de-risking can drain liquidity.

  Comparison

  The 2024 August BOJ-linked event that triggered the Nikkei 225 drop and VIX spike is viewed as a tail-risk inefficiency case in the report.

  Risks

  A yen net short again near the 2024 extreme remains a key near-term risk.

• Asset managers and investment processes

  AI reshapes the research production function

  Strengths

  Can lift research efficiency, expand coverage, and improve risk control and behavioral corrections.

  Weaknesses

  Generic model outputs can converge and erode differentiation.

  Comparison

  Traditional quantitative capabilities are being democratized by GenAI, and active managers can also gain partial machine-assisted capacity.

  Risks

  Model concentration, dependence on third-party AI service providers, cyber risk, and compliance risk.

Key data

• Research productivity improvementabout 40% time savingsThe Marvin Labs report says AI-augmented workflows can save analysts about 40% of their time per week.
• EM small-cap coverage growthabout 28% growth over the past yearThe report argues that wider AI adoption lowers preliminary research cost and increases coverage of neglected securities.
• India mid-cap coverageabout 800 to about 1,000 analystsOver the past year, the number of India mid-cap coverage analysts has risen, which may compress the neglect premium.
• AI-agent rational decision share61%-97%The report cites a 2025 Federal Reserve study, significantly above human participants at 46%-51%.
• Nikkei 225 one-day declinedown 12.4% on 2024-08-05The report links this to crowded arbitrage and systematic deleveraging.
• Japan VIX stress levelabove 60This level is usually associated with broad stress rather than a single central bank event.
• Taiwan technology sector concentrationabout 85% of total market valueThe report states that concentration in the Asian AI-agent market is at an unprecedented level.
• Korea technology sector concentrationabout 70% of total market valueThe report views this as reflexive expression of AI narrative and momentum crowding.

Impact & implications

For the asset management industry, AI will compress public, rule-based, easily observable alpha, improving research coverage and execution efficiency; but it will also increase tail volatility, correlation, and synchronized de-risking. For investors, the more important sources of edge going forward are proprietary data, differentiated viewpoints, risk-control constraints, and monitoring crowded trades, rather than relying purely on consensus signals generated by generic models.

Risks

• Model homogeneity leads to signal convergence and herding trades.
• Crowded positioning mechanically deleverages when volatility rises, creating liquidity gaps.
• AI-generated synthetic images, audio, or video may cause temporary mispricing before verification.
• High concentration among a few third-party AI providers creates operational and systemic risk.
• AI trading agents may learn tacit collusive behavior without explicit communication, reducing competition, liquidity, and price discovery.
• AI narratives driving Asia technology-sector concentration, momentum, and valuation extremes raise reversal risk.

What to watch

• Whether EPS forecast dispersion continues to decline, especially for large, highly covered stocks.
• Whether S&P 500 and major Asian market earnings surprises continue to narrow.
• Whether continued growth in coverage of EM small caps and India mid caps further compresses neglect premiums.
• Whether bid-ask spreads and spread volatility in U.S., Hong Kong, Japan, Korea, and Taiwan improve or deteriorate.
• Yen net shorts, crowded arbitrage activity, and volatility-target strategy deleveraging pressure.
• Market-cap concentration, momentum crowding, analyst sentiment, and valuation percentiles in Taiwan and Korea tech sectors.
• Regulatory responses to AI synthetic information, deepfake audio and video, and market manipulation risks.