Black box algo

Algorithmic trading means placing orders through automated, pre-set instructions. These instructions take account of price, timing and volume. SEBI has a framework for retail investors who use algo trading. It splits algos into two kinds, based on how much the user can see.

An execution algo, or white box algo, is fully open. Its logic is disclosed to the user. The user can copy it. Its decision rules are easy to see and understand.

A black box algo is the opposite. Its logic is not known to the user. It is not replicable by the user, so he cannot copy it. And the user cannot see how it works inside, or why it does what it does.

Here the user trusts a strategy he cannot check. So SEBI puts the burden on the provider. The black box algo provider must register as a Research Analyst. It must also keep a research report for every algo.

White box versus black box (Chapter 6, section 6.6):

Rules that apply to all retail algos under the framework (SEBI circulars of 4 February 2025 and 30 September 2025, stated in the workbook as proposed with effect from 1 October 2025):

• Orders per second. Up to 10 OPS is a low frequency algo: registration not required, generic algo ID. Above 10 OPS is high frequency: registration mandatory via the broker, unique exchange ID.
• Algo ID. The exchange assigns a unique identifier to each registered algo for surveillance.
• API access. No open APIs; access mapped to a broker-approved static IP; OAuth only and mandatory two-factor authentication.
• Family use. An algo may be shared within the family — self, spouse, dependent children and dependent parents only — with 2FA.
• Brokers deal only with empanelled algo providers after due diligence, and are solely responsible for handling investor grievances related to algo trading. SEBI and the exchange will not entertain any complaint.
• Kill switch — an emergency function to halt trading on predefined conditions if an algorithm malfunctions.

Separately, SEBI has directed brokers offering algo services not to refer to past or expected returns from any algo strategy, and not to associate with any platform that does.

An illustrative case built on the workbook's rules.

Suresh, a retail trader in Pune, is offered a subscription through his broker to an options strategy advertised as "Rs 50,000 a month on Rs 5 lakh capital". The provider will not explain how trades are chosen.

What kind of algo is it? Its logic is not disclosed and Suresh cannot replicate it — a black box algo.

What must be true before he can use it:

The red flag. The "Rs 50,000 a month" claim is a reference to expected returns. Brokers may neither make such a reference nor associate with a platform that does. On the workbook's figures that promise is a return of 50,000 ÷ 5,00,000 = 10% a month.

If it goes wrong. Suppose the algo loses Rs 1,20,000 in a week. Suresh's complaint goes to his broker, who is solely responsible for algo-related grievances. SEBI and the exchange will not take it up.

If the provider tweaks the strategy next month, the changed version must be registered as a new algo.

Chapter 6, section 6.6 (Algorithmic trading), defines algo trading, records SEBI's caution about unregulated platforms and return claims, and tabulates the retail algo framework, including the white box / black box categorisation. Expect questions on which type needs Research Analyst registration, what counts as a new algo, the 10 OPS threshold, and where algo trading complaints go.

• Black box = logic hidden, not replicable; provider registers as a Research Analyst. White box needs no registration.
• A logic change in a black box algo is a new algo — it must be registered again. Not so for white box.
• "Black box" is about transparency, not speed. The 10 OPS threshold decides low versus high frequency, separately.
• Algo trading complaints go to the broker. SEBI and the exchange will not entertain them.
• Brokers must not refer to past or expected returns of any algo strategy.
• Family sharing covers self, spouse, dependent children and dependent parents only.
• Check currency: the workbook describes the framework as proposed with effect from 1 October 2025.