Search for AI trading and you will find products promising consistent returns from machine learning. The honest position is narrower and less exciting: AI genuinely helps with execution — how to buy without moving the price — and helps far less with prediction, which is what everyone actually wants it for.
What Algorithmic Trading Is
Algorithmic trading means using computer programs to place orders according to defined logic. Most of it involves no AI at all — a rule that buys when a moving average crosses another is an algorithm, not machine learning.
| Rule-based algo | ML-based approach | |
|---|---|---|
| Logic comes from | A person’s stated strategy | Patterns inferred from data |
| Behaviour | Deterministic and inspectable | Harder to interpret |
| Adapts to new conditions | No | Somewhat, if retrained |
| Share of Indian institutional volume | Substantial majority | Smaller |
Where AI Genuinely Works
Execution. An institution needing to buy shares worth several hundred crore cannot place one order — it would move the price against itself. Models decide how to slice the order across time and venues to minimise market impact. The prediction here is short-horizon and mechanical: how will liquidity behave over the next few minutes? That is a tractable problem, and it is where most institutional AI value sits.
Market making. Continuously quoting both sides and managing inventory risk. Fast, repetitive, well-defined.
Signal processing. Extracting structure from large volumes of data — order flow, filings, cross-asset relationships — as one input among many rather than as a decision.
Risk management. Real-time position monitoring, exposure limits and automated hedging.
Notice the pattern: these are operational applications with short horizons and clear feedback. None of them is “predict which stock will rise”.
Why Prediction Resists AI
Four structural reasons, and they are not solved by better models.
The signal is faint. Prices already reflect widely known information. Whatever predictive signal exists is small relative to noise, and finding it is genuinely hard.
The target moves. A credit model’s borrowers do not restructure their behaviour to defeat it. Markets do exactly that — a discovered edge attracts capital until it disappears. This is drift caused by the model’s own success.
Data is scarcer than it appears. Twenty years of daily Nifty data is about five thousand observations — a small dataset for a model with many parameters, and it contains only a handful of genuine market regimes.
Backtests flatter. Test enough strategies against history and some will look excellent through chance alone. This is the defining failure mode of quantitative trading, and it is why an impressive backtest is close to meaningless without out-of-sample validation.
Retail AI Trading Products
A direct view: most retail AI trading products do not deliver what they imply.
| Claim | Reality |
|---|---|
| “AI-powered consistent returns” | If it worked reliably, it would be run with proprietary capital rather than sold by subscription |
| Impressive backtest | Trivially produced by testing many strategies and showing the survivor |
| Verified past performance | Usually simulated, not actual traded results |
| “Used by institutions” | Institutional use is mostly execution, not the prediction being sold |
The economic argument is the strongest one. A genuinely profitable, scalable trading edge earns far more deployed as capital than sold as a subscription. Products marketed to retail investors are, by their business model, telling you something about their confidence.
See deepfake fraud in finance — fabricated endorsements of trading products are now common, and the Rule of 72 is a useful three-second filter on any returns claim.
The Regulatory Position
SEBI regulates algorithmic trading in Indian securities markets. Broad features:
- Algorithms routed through brokers require exchange approval
- Risk controls and order-to-trade limits apply
- Audit trails must be maintained
- SEBI has progressively tightened requirements around retail algo access and the marketing of strategies
The regulatory direction has been toward greater scrutiny of retail algo products, partly because of performance claims. Verify the current position on the SEBI website before acting — this area has moved repeatedly.
What This Means for an Individual Investor
- You are not competing with institutions on speed. That contest was settled by co-location and infrastructure you do not have.
- Your genuine advantage is time horizon. You can hold for a decade. A fund reporting quarterly cannot.
- Treat AI trading claims as claims. Ask for audited live results, not backtests.
- Cost compounds against frequency. Brokerage, STT and slippage on every trade erode returns in a way that patient investing avoids.
Key Takeaways
- Most algorithmic trading is rule-based, not AI
- AI genuinely helps with execution, market making and risk — short horizons, clear feedback
- Price prediction resists AI for structural reasons better models do not fix
- A discovered edge disappears as capital pursues it
- Financial time series are smaller datasets than they appear
- Backtests are trivially flattered by testing many strategies
- A genuine edge earns more deployed than sold by subscription
- Your real advantage as an individual is time horizon, not speed
Frequently Asked Questions (FAQ)
Q: Does AI work for stock trading?
It works well for execution — deciding how to place large orders without moving the price — and for market making and risk management. It works far less well for predicting which stocks will rise, which is what most retail AI trading products claim to do.
Q: Why can’t AI predict the stock market?
Because prices already reflect known information, any edge disappears as capital pursues it, financial time series contain fewer genuine observations than they appear to, and backtests are easily flattered by testing many strategies. These are structural constraints, not modelling limitations.
Q: Are AI trading bots worth buying?
Approach with strong scepticism. A genuinely profitable and scalable trading edge earns far more deployed as capital than sold by subscription. Ask for audited live trading results rather than backtests, and treat impressive historical curves as close to meaningless without out-of-sample evidence.
Q: Is algorithmic trading legal in India?
Yes, and it is regulated by SEBI. Algorithms routed through brokers require exchange approval, risk controls and order-to-trade limits apply, and audit trails must be maintained. SEBI has progressively tightened rules around retail algo access — verify the current position on sebi.gov.in.
Q: What is the difference between algo trading and AI trading?
Algorithmic trading means placing orders by program according to defined logic, most of which involves no AI. AI trading uses models that infer patterns from data rather than following stated rules. The substantial majority of Indian institutional algo volume is rule-based.
Q: Why do backtests look so good but live results disappoint?
Because testing many strategies against history guarantees some look excellent by chance. Unless a strategy is validated on data never used in its development, an impressive backtest mainly demonstrates that enough variations were tried.
Q: What advantage does an individual investor actually have?
Time horizon. You can hold a position for a decade without reporting quarterly performance to anyone. Institutions cannot. Competing on speed against firms with co-located infrastructure is not a contest worth entering.
Related Reading: