Algorithmic trading, and who you are actually trading against
The answer
Most of the volume in modern markets is placed by computer programs, and the fastest of them react in microseconds — millions of times quicker than a person can move a finger. They are not predicting the future. They are winning a race you cannot enter, and the useful conclusion is about where not to compete.
Why this costs you money
A company announces results at 11:04. You read the headline at 11:04 and 20 seconds. You place your order at 11:04 and 35 seconds.
By then, several hundred programs have read the same announcement directly from the exchange feed, calculated the surprise against expectations, and traded. The price you are buying at already contains the news. You are not early. You are the last person in a queue that finished before you knew it existed.
This is the most expensive misunderstanding a retail trader can hold: the belief that reacting quickly to public information is a strategy. It was a strategy in 1995. It has not been one for 20 years.
The second cost is subtler. Because so many orders are placed and cancelled by machines, the order book you are watching is largely noise. Traders spend years learning to "read the tape" — the large buy order that appears and vanishes, the wall of sellers at a round number. Much of that is a program managing inventory, and some of it, historically, was placed specifically so that somebody would read intent into it.
How it works
"Algorithmic trading" covers 4 very different activities, and lumping them together is why the subject is so badly understood.
| Type | What it does | Timescale |
|---|---|---|
| Execution algos | Break a large institutional order into small pieces so it does not move the price | Hours |
| Market making | Quote both sides continuously, earn the spread, manage inventory | Milliseconds |
| Statistical arbitrage | Trade relationships between instruments that have drifted apart | Minutes to weeks |
| Latency arbitrage | Act on a price change at one venue before it is reflected at another | Microseconds |
Only the last one is a pure speed race, and it is a small part of the volume. Most algorithmic trading is not a genius machine forecasting the market. It is a fund manager's decision, made by a human over weeks, being carried out carefully by a program so that the market does not notice.
Speed comes from 3 places. Firms rent space for their servers inside the exchange's own data centre — this is called co-location, and it removes the distance the signal has to travel. They subscribe to the raw exchange data feed rather than a consolidated one. And they optimise their software down to the microsecond.
None of that is illegal. Exchanges sell co-location to anybody who pays. The point is not that it is unfair; the point is that it exists, it is expensive, and you are not buying it.
What it tells you, and what it does not
The dominance of machines tells you that the short-term, information-reaction game is taken. Anything that depends on being quick to a public fact has been industrialised.
It does not tell you the market is rigged against you. That framing is comforting and it leads nowhere. Spreads are narrower and costs are lower for retail investors today than in any previous era, and much of that is because of electronic market making.
And it does not tell you the machines are right. Algorithms have no view on whether a company will be worth more in 7 years. Almost none of them are trying to have one. They are competing over the next few seconds, which is precisely the reason the next few years are still available.
The decision rule
Never compete on a dimension where you are structurally the slowest participant in the room.
That rules out 4 things:
- Trading in the first 15 minutes on overnight news.
- Trading a headline you read on an app or a channel.
- Market orders — which are a promise to pay whatever the fastest
participant decides.
- Reading intent into individual orders in the depth window.
And it leaves the 2 things a retail investor genuinely has that a high-frequency firm does not: a long time horizon, and no obligation to trade at all.
A fund must deploy its capital. A market maker must quote. You can sit still for a year, and nobody will ask you why. That is not a consolation prize. It is the only structural advantage available to you, and most people trade it away.
Try this now
Your own trade history contains the answer, and you have never looked at it this way.
- Open your broker's Reports, Tradebook or P&L section and download the last 6 months of trades. Most brokers offer a spreadsheet.
- Sort by order time.
- Count how many of your trades were placed between the open and 15 minutes after it. Note what fraction of your total that is.
- Now add up the profit and loss of just those trades, and compare it with the profit and loss of everything else.
What you should see. For most active traders, the first-15-minute trades are a large share of the count and a disproportionate share of the losses. Not always — but if it is true for you, you have just found a rule you can follow tomorrow that requires no skill, no research and no discipline beyond waiting.
If you have no trade history yet, do this instead. Take any stock that reported results in the last month. Set the chart to 1 minute and find the minute the announcement landed. Look at how much of that day's entire move happened in the first 2 minutes. That is the part you were never going to catch.
Three real cases
1. Knight Capital, 1 August 2012 (United States) — speed cuts both ways A faulty software deployment caused Knight Capital's systems to send a flood of unintended orders into the market. In roughly 45 minutes it accumulated losses of around $440 million — more than the firm had earned in revenue the previous quarter, and more than the cash it had on hand. Knight was recapitalised days later by an investor group that took most of the company, and merged away the following year. A machine that can trade in microseconds can also lose in microseconds, and the operator cannot get there in time to stop it.
2. Navinder Sarao (United Kingdom / United States) — the book was lying A trader operating from a house in west London used software to place large orders he did not intend to execute, cancelling them before they filled, to create a false impression of supply. He was arrested in 2015 and later pleaded guilty to spoofing and fraud charges in the United States. The relevant lesson for a retail trader is not the crime. It is that the large orders visible in a depth window are not always offers to trade — and anybody whose method depends on believing them is depending on something that can be manufactured.
3. NSE co-location (India) — access is not equal, and it is disclosed SEBI passed orders in 2019 concerning preferential access to the exchange's tick-by-tick data feed at the NSE co-location facility. On appeal in 2023 the Securities Appellate Tribunal set most of that order aside, finding a failure of due diligence rather than a regulatory violation, and the matter was later settled. Set aside the specifics of the case. The structural point stands in every market: the participants sitting inside the exchange's data centre receive information before the participant sitting at home, and this is a published, purchasable arrangement, not a secret one.
The question that resolves it
A novice asks: how do I get faster?
An expert asks: what game is being played on a timescale where speed does not decide the winner?
You will not win a race measured in microseconds. Nothing prevents you from winning one measured in years, and the participants who beat you in the first race are mostly not entered in the second.
What would make this wrong
If algorithmic trading made markets unbeatable, then no long-horizon investor would outperform after 2010. Plenty have, in every market.
The honest limits.
Some algorithmic strategies do operate over weeks and months — trend following, factor investing, systematic value. These compete directly with a patient investor and they are formidable. "Machines only do short-term" is a simplification.
And electronic markets have genuinely improved things for ordinary investors. Spreads are a fraction of what they were, commissions have collapsed, and execution is more reliable. Anybody nostalgic for the trading floor is nostalgic for wider spreads paid by people exactly like you.
The correct conclusion is narrow and it is not despair. It is: do not enter races you cannot win, and notice which race you are entering.
In India
Algorithmic trading is a large and growing share of volume, and it dominates the derivatives market, where the majority of India's trading activity now sits.
SEBI regulates it more tightly than most markets. Algorithms used by brokers require exchange approval, and SEBI has moved to bring retail-facing algo platforms — the ones sold to individuals with promises of automated profits — under a formal framework.
Two Indian specifics matter to an individual.
- Co-location is available at both exchanges and is used heavily. The distance between a home internet connection in a tier-2 city and a server rack in the exchange's data centre is measured in milliseconds, and milliseconds are a geological age here.
- Expiry-day activity in index options concentrates an extraordinary amount of automated trading into a few hours. SEBI's own studies of individual traders in the equity derivatives segment have repeatedly found that the large majority lose money. That is the clearest evidence available of what happens when retail participants compete on the machines' preferred timescale.
In the United States
High-frequency firms account for a very large share of equity volume, and the infrastructure is mature: co-location at every major venue, microwave links between Chicago and New Jersey built to shave microseconds off the path, and a regulatory framework under Regulation NMS that ties venues together.
Two features are specific to the US.
- Payment for order flow. Retail orders are commonly routed to wholesalers rather than exchanges. The retail investor usually gets a slightly better price than the public quote; the wholesaler gets a flow of orders that is, on average, less informed than the flow arriving at an exchange.
- Off-exchange venues. A large share of volume executes where quotes are not displayed beforehand.
Individuals in the United States can also build and run their own automated strategies with far fewer restrictions than in India, through broker APIs. This is freedom, and it is also a great deal of rope.
Where they differ, and what that tells you
India restricts what an individual may automate. The United States mostly does not.
What that tells you is the shape of the risk in each place. In the United States, an individual can build an automated strategy and lose money quickly through their own errors — the Knight Capital problem, at household scale. In India, the individual is more protected from that, but is also unambiguously the slowest participant in the room, with no route to competing on speed even if they wanted one.
The conclusion is the same in both countries and it is the reason this article exists: your advantage was never going to be speed. In India that is enforced by structure; in America it is discovered by experience. Either way, the strategy worth building is one where waiting is an asset rather than a handicap.
Carry this
- The information-reaction game is taken. Do not enter it.
- Your 2 real advantages: a long horizon, and no obligation to trade.
- No market orders. No first 15 minutes. No headline trades.
- Large orders in the depth window are not promises.