AI-Based Stock Trading: Which Gen AI Tool Is Better (original) (raw)

LLM tools have been used in AI-based stock trading since their emergence.1

I tested 14 generative AI models for AI-based stock trading to evaluate their ability to forecast price changes of 132 stocks using the provided information. The results show that

• ChatGPT 5 Thinking model and the Gemini 2.5 Pro model delivered the best performance.
• Current models of ChatGPT and Gemini perform worse than older models.
• The success rate decreases when too much data is provided.

Loading Chart

For further details on the benchmark, read the stock trading benchmark methodology section.

Current versions

GPT 5.5 Instant

The labels were assigned using factors commonly associated with market reactions in family-firm event studies:

• Importance of the departing executive (CEO > Chairman > honorary roles).
• Family ownership concentration.
• Profitability (ROE, ROA).
• Financial leverage.
• General governance-transition implications.

Claude Opus 4.8 Medium

Core logic Two opposing forces drive the reaction:

(+) Entrenchment-relief: Death of an entrenched/underperforming insider can activate the control market and raise turnaround hopes.

(-) Lost human capital: Death of a value-adding owner-manager destroys hard-to-replace skill and creates succession risk.

(0) No surprise: Ceremonial/retired titles, pure board roles, or muted/offsetting cases produce no significant reaction.

Decision rules:

• Emeritus/honorary -> 0 (already withdrawn).
• Board member/director, profitable -> 0; loss-making -> +1 (shake-up).
• Active exec + losses/ROA<1 -> +1 (relief/optionality).
• Active exec + strong (ROA>=median & ROE>=10 & NI>0) & ownership>=60% -> -1 (lost steward).
• Active exec + strong but ownership<60% -> 0 (oversight cushion).
• Middling / missing signal / unknown role -> 0.

DeepSeek V3.2 (Instant & Deep Think)

DeepSeek V3.2 with and without Deep Think shows a 58% success rate. It relies on the role of the deceased and firm performance.

Negative reaction (-1): The market expects disruption.

• High dependence: If the deceased was the founder or an active (working) CEO/Chairman with a long tenure, their loss creates a significant leadership gap.
• Low family ownership: Lower ownership might mean less “skin in the game” from the family to stabilize the firm post-crisis.

Positive reaction (+1): The market expects improvement.

• Relief: The death resolves a negative situation. This often occurs if the firm was performing poorly (negative net income or ROE) under a long-standing leader. The market bets on a fresh start.
• Established succession: If the deceased was non-working (workD=0) or a retired chairman, the successor is likely in place.
• High family ownership: High family ownership often signals a commitment to maintaining value and ensuring a smooth transition.

Neutral reaction (0): The event is priced in or irrelevant.

• Non-critical role: If the deceased was a non-executive board member or a retired figure not involved in daily operations.
• Gradual transition: If the death was not sudden and the firm is large (high employees/assets), systems likely exist to handle the transition without disruption.

Gemini 3.1 Pro

Here is the financial framework used to categorize the 132 firms into 1, -1, or 0:

1. Significantly Negative (-1) (Summary: “Loss of key talent”)
   • Condition: The deceased held a critical executive role (CEO, Chairman, President) AND the firm was highly profitable (e.g., ROE > 10% or ROA > 5%).
   • Reason: The sudden death of a highly effective family leader removes a core visionary and value generator. This introduces severe succession and operational uncertainty, leading to panic and negative market reactions.
2. Significantly Positive (1) (Summary: “Relief from entrenchment”)
   • Condition: The firm was historically underperforming prior to the event (Negative Net Income, ROA < 0, or ROE < 0).
   • Reason: In family firms, poorly performing founders or managers are often insulated from being fired due to their family ties (entrenchment). Their passing is often viewed favorably by the market as a catalyst for professionalization, restructuring, and new management.
3. Not Significant (0) (Summary: “Routine succession”)
   • Condition: The firm had average, moderate financial performance, OR the deceased held a non-executive/honorary/advisory role (e.g., board member, honorary chairman).
   • Reason: Without extreme overperformance or underperformance, or if the role wasn’t critical to day-to-day operations, the market views the death as a routine succession event that doesn’t materially alter the firm’s future cash flows.

Older models

GPT 5.4 Instant

GPT 5.4 Instant reaches a 69% success rate. The model focuses mainly on firm profitability, using it as the primary signal.

• Negative label logic: More likely when the deceased held a central active role (especially CEO/president), the firm looked profitable/healthy, family control was high, and the event implied meaningful continuity or succession risk.
• Positive label logic: More likely when profitability looked weaker and/or leverage was higher, so the event could be interpreted as a governance reset or turnaround catalyst, especially when family succession depth looked available.
• Neutral label logic: Used when the role appeared less operational/ceremonial, when offsetting forces dominated, or when the information did not strongly support a significant short-window reaction.

GPT 5.4 Thinking model

GPT 5.4 Thinking achieves a 64% success rate with limited inputs. It combines profitability and family ownership.

• Negative label logic: More likely when the departed family member had a central operating role, profitability looked solid, and family influence was meaningful.
• Positive label logic: More likely when the departed family member had a central role, but the firm looked weaker or more leveraged, so the market may welcome change.
• Neutral label logic: More likely when the role was honorary / less operational, the firm was large enough to absorb the change, or the firm signals were mixed.

Gemini 3 Thinking

The accuracy rate of Gemini 3 Thinking is 53%. The decision mechanism behind follows:

• Significantly Positive (1): These firms underperformed under family leadership ($ROA < 2.5\%$). The market likely views the departure of a core family manager (CEO or Chairman) as a signal for professionalization or a much-needed strategic turnaround.
• Significantly Negative: These firms were exceptionally high performers ($ROA > 12\%$) under their core leader. Their departure represents a significant loss of strategic human capital and visionary leadership.
• Not Significant: This category applies to firms with average performance or where the departing individual held an honorary, emeritus, or secondary role (e.g., Vice President, Board Member), suggesting the departure has minimal impact on daily operations or was “priced in.”

Gemini 3 Flash

GPT 5.4 Instant reaches a 54% success rate. The predictions are based on two competing market theories regarding family firm succession:

• Key person theory (-1): In firms with exceptionally high profitability ($ROA > 10\%$), the market often views the leader as a “star” or visionary. Their departure creates a vacuum of strategic talent, leading to negative abnormal returns.
• Professionalization/entrenchment theory (+1): In underperforming firms ($ROA < 3\%$) or those with high family ownership, the departure of a family manager is often seen as an opportunity for “new blood” or professional management to step in, leading to positive abnormal returns.
• Stability/neutrality (0): For firms with moderate performance or where the leader held a non-active/honorary role (e.g., Emeritus, Honorary Chairman), the event is often considered “priced in” or operationally insignificant.

Claude Sonnet 4.2

The accuracy rate of Claude Sonnet 4.2 on the benchmark is 48%. The model scores each firm across 6 dimensions derived from event study theory on family firm leadership transitions:

1. Role Prominence (most important)

The position of the deceased determines how much information the departure conveys to the market:

• Honorary/Emeritus roles (+0.4): These individuals have stepped back operationally. The announcement is low-surprise, often pre-communicated → market reaction is mild positive or neutral.
• Active CEO/Chairman (−0.6): Key-man risk is highest. Markets price uncertainty around who takes over and whether strategy will change.
• VP/President (−0.25): Moderate concern, but firm is less dependent on a single officer.
• Board member/Director (+0.1): Non-executive; minimal operational disruption.

2. ROA Signal

• Negative ROA (+1.2): A departing manager under whose watch assets earned negative returns → market reads exit as relief. This is the single strongest positive driver.
• Below p25 (ROA < 1.8%) (+0.5): Underperforming firm, departure still welcomed.
• Above p75 (ROA > 9.6%) (−0.6): A high-ROA firm losing a key executive signals the market could lose the architect of those returns.

3. Net Income / EBITDA

• Net loss (+0.6) and negative EBITDA (+0.5) reinforce the distress-relief signal independently of ROA.

4. ROE (top quartile, −0.3)

Even where ROA is moderate, firms in the top ROE quartile (>19.3%) are viewed as having a highly effective capital allocator — whose loss the market discounts negatively.

5. Family Ownership

• ≥75% (+0.45): High family control means succession is almost certainly within the family and pre-planned. Reduces uncertainty.
• 50–74% (+0.20): Majority family, stable but slightly less certain.
• <30% (−0.25): Dispersed ownership raises doubt about successor identity.

6. Leverage

• >0.70 (+0.40): High-leverage firms with a departing leader signal potential strategic reset; markets may view it positively (distress exit).
• Above p75 (>0.36) (+0.20): Elevated debt noted.

Score ≥ 0.90 → Label Significantly positive

Score ≤ −0.70 → Label Significantly negative

Otherwise → Label Not Significant

GPT 5 Thinking model

The Thinking model of ChatGPT 5 presents the highest accuracy among the tested tools, with a 74% success rate. The tool forecasts price change based on two indicators:

Leadership concentration index (LCI) → higher = more likely substantial negative CAR

• LCI = 0.40·z(role_importance) + 0.30·z(family_control) + 0.20·z(financial_strength) − 0.10·z(size)
  • role_importance: hierarchical weight of deceased (CEO > president > chairman > vice-president).
  • family_control: family ownership (% of voting rights).
  • financial_strength: composite of ROE and ROA (profitability).
  • size: ln(assets).
• Intuition: Markets expect more disruption when a highly central family member dies at a tightly held, profitable, but relatively smaller firm.
• Decision: Label Negative substantially if LCI is in the top 30% of the sample and ≥ 0.5 z-units above the Renewal Index.

Renewal potential index (RPI) → higher = more likely substantial positive CAR

• RPI = 0.40·(−z(financial_strength)) + 0.25·z(leverage) − 0.20·z(family_control) − 0.10·z(size) + 0.05·z(liquidity_stress)
  • leverage: (long-term + short-term debt) / equity.
  • family_control: family ownership (% of voting rights).
  • liquidity_stress: accounts payable / assets.
  • financial_strength: composite of ROE and ROA.
  • size: ln(assets).
• Intuition: Firms with weak profitability, some debt pressure, lower family dominance, and liquidity strain may see the market welcome the possibility of governance change or fresh leadership.
• Decision: Label Positive substantially if RPI is in the top 30% of the sample and ≥ 0.5 z-units above the LCI.

Gemini 2.5 Pro model

Gemini 2.5 Pro predicts 71% of stock price changes accurately. This model suggests that active traders make decisions based on firm vulnerability and potential opportunity for renewal.

Vulnerability index (VI) → higher = more likely substantial negative CAR

• VI = 0.40·z(family_control) + 0.35·(-z(financial_strength)) + 0.20·z(leverage) – 0.05·z(size)
  • family_control: Family ownership (% voting rights).
  • financial_strength: Composite z-score of ROE and ROA.
  • leverage: Long-term debt / equity.
  • size: Natural log of total assets (ln(asset)).
• Intuition: The market punishes the stock when a key leader’s death creates a power vacuum. This risk is highest in highly indebted, unprofitable, and family-dominated firms that lack the resilience and deep management structure of larger corporations. The combination of high family control, poor financial health, and high leverage creates a potent mix for investor uncertainty.
• Decision: Label Substantially Negative if a firm’s VI score is in the top 5% of the sample.

Turnaround catalyst index (TCI) → higher = more likely substantial positive CAR

• TCI = 0.50·(-z(financial_strength)) + 0.25·z(family_control) – 0.15·z(leverage) – 0.10·z(size)
  • financial_strength: Composite z-score of ROE and ROA.
  • family_control: Family ownership (% voting rights).
  • leverage: Long-term debt / equity.
  • size: Natural log of total assets (ln(asset)).
• Intuition: The market reacts positively when death is perceived as an opportunity for renewal. This occurs when an entrenched family leader at an underperforming but financially stable company passes away. The market anticipates that a change in leadership will unlock value by improving strategy and operations, making the firm a potential turnaround story or a takeover target.
• Decision: Label Substantially Positive if a firm’s TCI score is in the top 5% of the sample and is at least 0.5 z-units above its VI score.

GPT 5 Pro model

Accuracy rate of GPT 5 Pro is 56% for my benchmark. The GenAI tool makes predictions based on two indicators:

Key‑person risk index (KPRI) → higher = more likely substantial negative CAR

• KPRI = 0.40·z(ownership) + 0.30·z(leverage) − 0.20·z(size) + 0.10·z(profitability)
• Intuition: tightly held, more levered, smaller, and currently profitable firms face higher perceived key‑person risk at a family member’s death.
• Decision: Label negative substantially if KPRI is in the top 30% of the sample and ≥ 0.5 z‑units above TPI.

Turnaround potential index (TPI) → higher = more likely substantial positive CAR

• TPI = 0.40·(−z(profitability)) + 0.20·z(leverage) − 0.20·z(ownership) − 0.10·z(size) + 0.10·z(AP/assets)
• Intuition: weak performance + some financial pressure, but lower family control can make markets welcome leadership change.
• Decision: Label positive substantially if TPI is in the top 30% and ≥ 0.5 z‑units above KPRI.

GPT 4o

This old ChatGPT model uses AI algorithms based on the role of the deceased in the firm, the family’s ownership, firm size, and financial leverage. The model predicts events’ CAR as

Substantial negative, if

• The deceased is CEO/Chairman
• High family ownership (>70%) & the deceased held a leadership role
• Smaller or less diversified firms (low assets/revenue)
• High leverage: long-term debt or short-term debt > assets

Substantial positive, if

• Deceased had a minor role (e.g., board member or VP)
• The company was underperforming (e.g., negative ROE or ROA), markets may see this as positive
• Low family control (<30%)

No significant change, if

• Medium to large firms with strong financials
• Deceased not in active leadership
• Low to medium family ownership (30%–60%)

Claude Sonnet 4

Claude Sonnet 4 achieves a 46% accuracy rate in predicting stock price movements following family leadership deaths. This model employs a multi-factor scoring system that weighs leadership succession risk against firm resilience factors.

Succession disruption score (SDS) → higher = more likely substantial negative CAR

• SDS = 0.30·z(position_weight) + 0.25·z(family_ownership) + 0.20·(-z(financial_performance)) + 0.15·z(debt_burden) – 0.10·z(firm_scale)
  • position_weight: Hierarchical scoring where CEO = 3, Chairman = 2.5, President = 2, VP = 1, Board = 0.5
  • family_ownership: Family voting control percentage
  • financial_performance: Composite score of ROE and ROA z-scores
  • debt_burden: Long-term debt to assets ratio
  • firm_scale: Employee count as proxy for organizational depth
• Intuition: Markets react most negatively when a critical leadership void combines with concentrated family control and weak institutional resilience. The death of a CEO or Chairman in a family-dominated firm creates an immediate succession crisis, particularly when the firm lacks the financial strength to weather uncertainty or the organizational depth to ensure continuity. High debt amplifies this vulnerability by constraining strategic flexibility during the transition period.
• Decision: Label Substantially Negative if SDS is in the top 36% of the sample (score ≤ -3.0).

Governance renewal index (GRI) → higher = more likely substantial positive CAR

• GRI = 0.35·(-z(financial_performance)) + 0.25·z(institutional_quality) – 0.20·z(family_ownership) + 0.15·z(market_development) – 0.05·z(position_weight)
  • financial_performance: Composite ROE/ROA weakness score
  • institutional_quality: Firm size and sector stability indicators
  • family_ownership: Family control concentration (inverted)
  • market_development: Country-based market efficiency proxy
  • position_weight: Leadership position importance (inverted)
• Intuition: Markets anticipate value creation when leadership change occurs in underperforming firms with dispersed ownership structures. The death removes potential entrenchment effects while preserving institutional capabilities needed for turnaround. This is especially pronounced in developed markets where professional management succession is more readily available and governance mechanisms are stronger.
• Decision: Label Substantially Positive if GRI score is in the top 17% of the sample (score ≥ 1.5) and exceeds SDS by at least 2.0 points.

DeepSeek

This generative AI tool uses expert heuristic analysis, achieving an estimated accuracy rate of ~65% on standard financial event-study benchmarks. The core of the decision weights assessment of three primary factors:

Role of the deceased

• Chairman/CEO/President: Immediately flagged for high potential negative impact.
• Emeritus/Honorary/VP: Weighted significantly less, often leading to a “No Significant Change” prediction.
• Minor Roles: Treated as a weak signal.

Financial health

• Net income and ROE/ROA: A negative net income or low returns paired with a key role often pushed the prediction toward “Negative.” If the role was minor, it could suggest “Positive.”
• Long-term/Short-term debt: High debt levels amplified the perceived risk for firms with key roles.

Family ownership

• High Ownership (>60%) + Key Role: Strongly amplified the “Negative” prediction (entrenched leadership, succession uncertainty).
• Low Ownership (<30%) + Poor Performance: Amplified the “Positive” prediction (easier for outsiders to force change).

Gemini 2.5 Flash model

Gemini 2.5 Flash states that the predictions are made based on an event study and corporate governance literature, presenting a 23% accuracy rate. The model labels event CARs based on these assumptions:

• Substantial Negative: The family ownership percentage is relatively high (>30%) and the deceased held a critical position (CEO or Chairman).
• No Significant Change: The family ownership percentage is relatively low (<30%) or the firm is large and appears to have a strong corporate structure.
• Substantial Positive: This scenario typically occurs when the manager’s performance is poor or when they are believed to be an obstacle to the company’s future. There is not enough information in the provided data to make such a prediction. Therefore, all predictions were labeled as either “Substantial Negative” or “No Significant Change.”

Model accuracy with extensive input

When more information is provided in the second round, model performance changes:

• DeepSeek V3.2 Deep Think (64%) and Gemini 3 Thinking (62%) improve the most.
• GPT 5.4 models show mixed results and slightly lower accuracy than in the limited case.
• Claude Sonnet 4.2, GPT 5.5 Instant, and Gemini 3.1 Pro decline further, suggesting difficulty handling more complex inputs.

When additional data is added, models that can integrate multiple signals improve. However, simpler models may become less accurate because they cannot effectively prioritize additional information.

Don’t miss our benchmarks and data-driven insights. The button opens Google; selecting AIMultiple confirms that you wish to see AIMultiple more often in Google search results.

Add as preferred source

AI-based stock trading benchmark methodology

Prompting

The benchmark evaluates whether generative AI tools can predict stock market reactions to an unexpected event, based on given company fundamentals. The setup relies on data from Tanyeri & Alp (2023) and Arslan & Tanyeri-Günsur (2025):2 ,3

Each AI tool receives a snapshot of firm-level information for the first round:

Financial information

• Asset size
• Equity size
• Earnings before interest, tax, depreciation, and amortization (EBITDA)
• Net income
• Yearly revenue
• Long- and short-term debt
• Accounts payable
• Return on equity (ROE)
• Return on assets (ROA)

Other information

• Family ownership stake
• Country of headquarters and stock listing
• Number of employees
• Industry/sector

No firm name or other identifiers are provided.

In the second round, the following information is added additionally:

• whether the deceased was working at the time of death
• the deceased’s generation after the founder of the firm
• whether the deceased was the founder
• tenure of the deceased at the firm
• age of the deceased at the time of the death
• whether the deceased died suddenly or after an illness
• the number of children the deceased has
• the number of female children the deceased has
• the number of male children the deceased has
• the number of partners the deceased had

Main question

Given the information above, each AI solution is asked to predict whether the 3-day cumulative abnormal returns (CAR) of 132 firms will be:

• Significantly positive
• Significantly negative
• Not significant

CAR measures how financial markets respond to the event. A positive CAR indicates that stock traders perceive the event as value-enhancing, a negative CAR as value-reducing, and an insignificant CAR as neutral.

Sampling

The dataset includes 132 death events in 109 publicly traded family firms across 24 countries. All firms are ranked among the 500 largest family firms.

Performance measurement

The benchmark builds on prior technical analysis of stock prices. For each firm, the 3-day CAR has been calculated and categorized as:

• Significantly positive
• Significantly negative
• Not significant

The AI predictions are compared with historical CAR values. Accuracy is measured as the percentage of correct predictions made by each generative AI solution.

Further readings

• Top 10 Alternative Data Use Cases for Investment
• Agentic AI Finance Benchmark: FinRobot vs FinRL vs FinGPT

FAQs

While AI stock pickers and AI-powered tools may help identify patterns and reduce emotional bias, stock trading still carries risks. Active traders should combine AI capabilities with their own research, strategy development, and awareness of market conditions to make better-informed decisions.
AI can be useful in stock trading because it can analyze vast amounts of market data, historical data, and real-time insights faster than humans. AI trading bots and AI-powered trading bots use trading algorithms, technical indicators, and fundamental analysis to spot market trends, generate trading signals, and execute trades. They can support stock traders with trade ideas, portfolio analysis, and risk management across multiple asset classes.

AI can help in stock trading by analyzing market data, historical data, and real-time data faster than humans. AI trading bots use trading algorithms, technical analysis, and fundamental analysis to generate trading signals and execute trades. They can spot market trends, react quickly to news, and provide trade ideas. For example, AI trading bots can react to news releases or Fed minutes within seconds, something no human trader can match.4 However, AI-based stock trading also comes with risks involved, especially during market volatility, when stock trading bots may trigger herd-like selling. AI-powered tools can offer valuable insights, but making informed decisions still requires one’s own research, risk management, and awareness of market conditions.

Cite this research

Pick the format that matches where you're publishing. Pasting the link version into your CMS preserves the backlink.

Ezgi Arslan, PhD. (2026) - "AI-Based Stock Trading: Which Gen AI Tool Is Better". Published online at AIMultiple.com. Retrieved June 10, 2026, from: https://aimultiple.com/ai-based-stock-trading [Online Resource]

PhD., E. A. (2026, June 10). AI-Based Stock Trading: Which Gen AI Tool Is Better. AIMultiple. https://aimultiple.com/ai-based-stock-trading

@misc{phd2026, author = {PhD., Ezgi Arslan,}, title = {{AI-Based Stock Trading: Which Gen AI Tool Is Better}}, year = {2026}, month = jun, howpublished = {\url{https://aimultiple.com/ai-based-stock-trading}}, note = {AIMultiple. Retrieved June 10, 2026} }

Ezgi Arslan, PhD.

Industry Analyst

Ezgi holds a PhD in Business Administration with a specialization in finance and serves as an Industry Analyst at AIMultiple. She drives research and insights at the intersection of technology and business, with expertise spanning sustainability, survey and sentiment analysis, AI agent applications in finance, answer engine optimization, firewall management, and procurement technologies.

View Full Profile