Algorithmic trading is the practice of handing a written, unambiguous trading rulebook to a computer, so that the computer — not you — decides when an order is sent. It is a change in who executes, not a change in what works. The rulebook still has to have an edge; the machine only removes hesitation, delay and the small improvisations a human makes under pressure. That is genuinely valuable, and it is also the entire claim. A program cannot make a losing rulebook profitable, and it will lose money faster and more consistently than you would have managed by hand.
What does 'algorithmic trading' actually mean?
An algorithm is a finite list of instructions with no gaps. That is a mathematical term, not a marketing one. A trading algorithm is therefore a list of instructions complete enough that a machine, which cannot use judgement, can follow it to a decision every single time — including on the days the answer is 'do nothing'.
That completeness requirement is the hard part, and it is where most people discover their strategy was never a strategy. 'Buy strong stocks near support on good volume' is a sentence, not an algorithm. It contains four undefined terms. Until each of them becomes a number or a condition, no computer can act on it — and honestly, neither can you, consistently.
So the useful definition is narrow: algorithmic trading is what you get once your rules are specific enough to be executed without you.
How is systematic trading different from algorithmic trading?
These two words are used interchangeably in most Indian coverage, and they are not the same thing.
Systematic trading is about the decisions — you follow a written rulebook rather than a view. Algorithmic trading is about the execution — a program sends the orders. You can be fully systematic with a printed checklist and a broker's mobile app, and never write a line of code. You can also automate a set of rules you invented last week, which makes you algorithmic and reckless at the same time.
The order matters. Systematic comes first. Automation is a delivery mechanism you bolt onto a rulebook that already exists and has already been tested.
| Systematic trading | Algorithmic trading | |
|---|---|---|
| What it fixes | Inconsistent decisions | Inconsistent and slow execution |
| Rules written down | Yes — that is the definition | Yes, necessarily |
| Who sends the order | Usually you, by hand | A program, through a broker API |
| Code required | None | Yes, or a broker-hosted no-code builder |
| Can it run without you awake | No | Yes — which is the risk as much as the benefit |
| What it cannot do | Stop you overriding the rulebook | Make a rulebook worth following |
Note — If you are not yet systematic, automation is premature. Being Systematic Without Writing Code, the next practical topic in this module, is the honest starting point for most readers — and a large number of people never need to go further than it.
Is algorithmic trading the same as high-frequency trading?
No, and the conflation does real damage — it convinces retail traders that they are competing with co-located servers when they are trading a daily-close breakout.
High-frequency trading is one narrow corner of algorithmic trading, defined by its time scale and its infrastructure. HFT firms compete on microseconds. They co-locate hardware inside the exchange's data centre, pay for the lowest-latency feeds, and run strategies whose entire edge is being ahead of somebody else's order by a margin no human perceives. That is a capital and engineering contest, and a retail participant is not in it.
Almost nothing a retail algo trader builds lives there. A rule that evaluates on the daily close and places one order the next morning is algorithmic trading. It is not remotely HFT, and the fact that HFT exists is not an argument against it — the two are not competing for the same trade.
What does an automated strategy actually do, step by step?
Strip away the vocabulary and every retail algo, however sophisticated, is the same loop running on a timer.
What are the parts of that loop?
- 1
Get data
Prices arrive from your broker's feed or a data vendor. This step fails more often than beginners expect — a missed tick, a stale candle or an unadjusted price after a bonus issue quietly corrupts every decision downstream.
- 2
Evaluate the rule
The rulebook is applied to the data as it stands right now. The answer is one of three things: buy, sell, or — on the overwhelming majority of evaluations — do nothing.
- 3
Run the risk checks
Before any order is built: is this instrument already held? Is total open risk within its cap? Is the account inside its loss limit for the day? These checks sit between the signal and the order deliberately, because that is the only place they can stop a bad order.
- 4
Size the position
Quantity is arithmetic from the stop distance and the risk budget, exactly as in manual trading. Automation does not change this formula; it only stops you rounding it up because you feel confident.
- 5
Send the order
A call to the broker's API. It can be rejected for margin, for a price band, for a frozen quantity limit, or because the exchange has halted the scrip — and your code must handle every one of those, not assume a fill.
- 6
Reconcile and log
Read back what actually filled, at what price, and update your position from fills — never from orders you believed went through. Then write the whole decision to a log you can audit later.
Does automation give you an edge?
No. This is the single most important sentence on the page.
An edge is a reason your rulebook makes money after costs — a real, describable feature of how the market behaves, which your rules capture. Automation is a delivery mechanism. It changes the fidelity with which the rulebook is executed, and it changes nothing about whether the rulebook was worth executing.
What automation genuinely gives you is narrower and still worth having: identical execution across every occurrence, no missed signals while you were in a meeting, no size inflated because the last trade worked, and a complete record of what happened and why. Those are consistency benefits. Consistency amplifies whatever the rulebook already is.
What does automation take away?
Every honest account of this has to include the trade-offs, because the marketing never does.
- Discretion, including the good kind. You will watch your program take a trade you would have skipped, on a day when skipping it was right. You cannot keep the override and keep the consistency — that override is the exact thing you automated to remove.
- The excuse of inattention. A manual trader who oversleeps takes no trades. A program with a broken exit condition trades all day into a market it was never designed for.
- Simplicity. You have now added software, a data feed, a network connection, a machine that must stay awake, and API credentials — every one of which is a new way to lose money that has nothing to do with the market.
- Cheapness, at small size. Costs are per-trade and mostly fixed in structure. A rule that trades often can be sound before brokerage, STT, exchange charges, stamp duty, GST and slippage, and unsound after them.
- The comfort of ambiguity. A vague rule can never be proved wrong. A coded one is falsifiable, which is the point, and it is uncomfortable the first time your favourite idea fails its own test.
How does trade frequency change the cost arithmetic?
This is where a lot of retail automation quietly dies, so it is worth doing the arithmetic before the code.
Net edge per trade = Gross edge per trade − (round-trip costs + slippage)- Gross edge per trade — the average outcome of the rule before any friction. It is an unknown you estimate, never a promise.
- Round-trip costs — brokerage, STT/CTT, exchange transaction charges, SEBI turnover fees, stamp duty and GST on the applicable components, both legs.
- Slippage — the gap between the price your rule assumed and the price you actually got. It grows with size and shrinks with liquidity.
- Frequency multiplies the bracket, not the gross edge. Doubling how often you trade doubles the friction with certainty and doubles the edge only if the edge was real.
Example — Illustrative only, with round numbers and no real instrument: if a rule is worth ₹40 per share gross and friction is ₹15 per share round-trip, it clears ₹25. Take the same rule to a shorter timeframe where the move is ₹12 and friction is unchanged, and it now loses ₹3 every time it is right. Nothing about the logic changed — only the size of the move it is trying to capture relative to a fixed toll.
Who is retail algo trading actually for?
| Situation | Is automation the answer? | What to do first |
|---|---|---|
| You have a written rulebook you have followed by hand for months | Possibly yes | Automate the execution of that exact rulebook. Change nothing else at the same time. |
| You keep overriding your own rules | Not yet | The problem is the rulebook or your belief in it. Code will enforce a rule you do not trust, which is worse. |
| You are losing and want a system that fixes it | No | Automation scales the current outcome. Find out why the rulebook loses first. |
| You cannot watch the screen during market hours | Often yes | This is the strongest honest case for automation. Start with a rule that evaluates once a day. |
| You want to test an idea over years of data | Yes — but not live | That is backtesting, and it is a research task. It does not require you to trade a single rupee. |
| Someone sold you a strategy with a claimed return | No | Verify who is selling it and under what registration. A claimed return you cannot audit is not evidence. |
What do you actually need to run an algo in India?
- A written rulebook with no undefined terms — including the conditions under which it must do nothing.
- Historical data clean enough to test on, corrected for splits, bonuses and delistings.
- A broker whose API you are permitted to use for order placement. Several Indian brokers publish trading APIs; the module's Broker APIs in India topic covers how that access works mechanically.
- A machine that stays on and connected through the session — a small VPS, more often than a laptop.
- Order and position reconciliation, so your program's idea of what you own matches the broker's.
- A kill switch you can reach from your phone, and the habit of testing that it works before you need it.
- An understanding of the regulatory position, which is the very next topic and is not optional reading.
What invalidates the case for automating a strategy?
Automation is a decision like any other, and a decision you cannot invalidate is a belief. Here is what proves it wrong.
- The rulebook still contains a term you cannot express as a number or a condition. You have not finished writing it, and coding it will only hide the gap behind an arbitrary default.
- The tested edge does not survive realistic costs and slippage. This kills more retail algos than any market condition does.
- The rule only worked in the data you used to build it. If it fails on a period you never looked at, you fitted noise — covered in full in Overfitting & Walk-Forward Validation.
- You cannot state, in advance, what would make you switch it off. A system with no shutdown condition will be shut down emotionally, at the worst point.
- The strategy needs execution quality you cannot achieve. If it assumes fills at the touch in an illiquid scrip, the backtest was trading a market that does not exist.
- You would not follow the rulebook by hand. Automating something you do not believe in produces the same losses, unattended.
- You cannot monitor it. An unmonitored program is not a strategy; it is an open-ended instruction to your broker.
Watch out — You will be shown backtested equity curves, monthly return grids and 'win rates' by people selling algo products. Treat all of them as unusable unless you can see the raw trade list, the exact cost assumptions, the data window and the periods that were excluded. A curve without those is a picture, not evidence — and a promised or guaranteed return on a trading strategy is a claim no one is permitted to make to you.
What should you learn, and in what order?
The honest sequence through this module
- Understand the regulatory position before you write any code — the next topic.
- Become systematic without automating anything. Follow a written rulebook by hand.
- Learn the research workflow: hypothesis, test, kill, log. Most ideas must die.
- Only then learn enough Python to test an idea — not to build a platform.
- Understand your data before you trust it. Corporate actions and survivorship bias break more backtests than bad logic.
- Backtest properly, then attack your own result until you can explain why it is not overfitted.
- Size and risk-manage the system as a portfolio, not as a series of clever trades.
- Subtract every cost honestly. Then paper trade, then go live at a size that genuinely does not matter.
- Build the monitoring and the kill switch before the first live rupee, not after the first incident.
Key points
Pro tip — Before you write any code, run your rulebook by hand for twenty trading sessions and log every decision, including the days it said do nothing. If you overrode it even once, automation is not your next step — understanding why you overrode it is. This exercise costs nothing and it filters out most of the strategies that would have been expensive to automate.
Frequently asked questions
What is algorithmic trading in simple terms?
It is giving a computer a complete, unambiguous set of trading rules and letting it send the orders instead of you. The rules still come from a human. The computer contributes consistency and speed of execution, not judgement and not an edge — it will follow a poor rulebook just as faithfully as a good one.
Is algo trading legal in India?
Algorithmic trading is a recognised, regulated activity in Indian markets rather than a grey area. SEBI regulates it, the exchanges set operational requirements, and retail access is routed through registered brokers. Because the framework has been evolving and the operational detail sits in circulars, verify the current position on sebi.gov.in and on your exchange's website before you deploy anything — the next topic in this module covers the shape of the framework and what to check.
Is algorithmic trading profitable?
Automation is not a source of profit, so the question cannot be answered about algo trading as a category. Profitability depends on whether the underlying rulebook has a real edge that survives brokerage, taxes and slippage. Automating a strategy that loses money produces losses more consistently, which is why anyone quoting a return figure for 'algo trading' generally is describing a product, not a fact.
Do I need to know programming to trade algorithmically?
To build and run your own algorithm, yes — usually Python, at the level of loading data, applying conditions and calling an API, which is far less than software engineering. To be systematic, no code is needed at all. Some brokers also offer hosted rule builders that require no programming, and the trade-off there is that you are limited to the logic the platform exposes.
How much capital do I need to start algo trading?
There is no regulatory minimum for trading capital, but the arithmetic sets a practical floor: costs are largely fixed per trade, so a small account trading frequently gives most of any gross edge back as friction. Add the running cost of a server and any data subscription, and a frequently-trading algo needs materially more capital to be viable than a once-a-day one on the same rules.
What is the difference between algo trading and HFT?
HFT is a small, specialised subset defined by microsecond time scales, co-located hardware and infrastructure spend that is out of reach for retail participants. Most retail algorithmic trading evaluates on daily or intraday candles and places a handful of orders — it is algorithmic because a program executes it, not because it is fast, and it is not competing with HFT for the same trade.
Can I just buy a ready-made algo strategy?
Strategies are sold, and the questions to ask are the same ones that apply to any financial claim: who is offering it, under what registration, and can you audit the raw trade list, the cost assumptions and the data window behind any figure shown to you. Nobody is permitted to promise or guarantee you a return on a trading strategy, and a performance figure you cannot independently check is marketing rather than evidence.