Algorithmic Trading vs API Trading: Where the Line Falls
Algorithmic trading describes how the decision to place an order was made: a system settled it, not a person choosing that order in turn. API trading, meaning trading through an application programming interface (API), describes how the order reached the market: a programmatic interface carried it, not a screen somebody typed into. The two words answer different questions, they get used as though they were one word, and a firm can be doing either without the other.
Here is what sits underneath that, and it is a single idea rather than a list of differences. Two separate things happen to every order that ever reaches an Indian exchange. Something decided it, and something carried it. One of these two words describes the deciding and the other describes the carrying, so putting them side by side as though a firm picks one is like asking whether a parcel was addressed or delivered. Both happened. Deciding and carrying are different events, they can be arranged in any combination, and each one raises questions of its own.
In India each of the two terms is governed by its own named instrument. The requirements, control specifications, thresholds and limits themselves live in those instruments. Requirements of that kind are set in rules that get revised, and a number copied out of them gets quoted back with total confidence years after it stopped being right. The durable part is the shape of the distinction, the name of each rulebook, and the habit of going to the source on the day the answer is actually needed.
Each term can be defined on its own without reaching for the other, and once both are defined, every arrangement drops into one of four squares that a single question tells apart without once looking at the connection.
What is algorithmic trading, on its own terms?
Start away from markets altogether. Think about the thermostat in a cold storage room behind a sweet shop. Nobody stands there watching the dial. Somebody wrote a rule once, that the compressor starts below a temperature and stops above another one, and after that the rule decides. The owner of the shop still decided something, and what he decided was the rule, not the individual switching on at eleven at night. Nobody would say the compressor came on because the owner chose it that evening. He was asleep.
Algorithmic tradingOrder decisions made by a system rather than by a person taking each order in turn. stands in that position with respect to orders. A rule is written once, the rule watches something, and when the condition it watches is met, an order goes out without anybody choosing that order on that morning. Algorithmic trading describes the origin of the decision and nothing else, so the test is whether a person settled this particular order or whether a system did. Sumana Rege, head of technology at the invented broker Bhadra Securities Private Limited, can describe such a rule in one sentence, and the sentence never mentions cables, screens or connections. The sentence mentions what is watched, what triggers, and what goes out.
Sophistication does no work in that definition. A rule with two conditions in it is a system deciding. A rule with two hundred conditions in it is also a system deciding. Nothing in the definition asks how clever the rule is, how fast it runs, how much it cost or who wrote it. The Securities and Exchange Board of India sets requirements for firms in this position and the exchanges set operational requirements of their own, a substantial subject set out at sebi.gov.in, nseindia.com and bseindia.com and treated under algorithmic trading in India.
What is API trading, on its own terms?
The second term looks at something else entirely. Setting aside who decided anything, what matters is the delivery. A letter can go to the post office in somebody's hand, or it can go down a chute in an office building that empties into the same postbag. The words inside the envelope are identical either way. The path changed, and the path says nothing whatsoever about who wrote the letter or why.
API tradingOrders reaching the market through a programmatic interface rather than through a screen a person types into. describes the path. An order can be entered by a person into a trading screen, or it can be handed to a programmatic interfaceA connection built for one piece of software to hand instructions to another, without a screen in between., a connection built so that one piece of software can pass instructions to another without a screen in between. API trading describes the route the order travelled and nothing else, so the test is what carried the order, not what produced it. When Bhadra Securities Private Limited offers such a connection to a client, the firm is offering a way in. The order the client sends through the connection, and whatever decided to send it, are separate matters entirely, and the connection being there settles neither.
API trading is easily mistaken for a description of behaviour. It is a description of plumbing. Plumbing is worth regulating, and the questions it raises are real ones, and none of those questions is answered by knowing what the water was for.
Both terms are now defined on their own. Which of the two describes how the order reached the market?
What is the actual difference between the two terms?
The whole distinction fits in one line, and everything after it is elaboration. One word describes the decisionWhat determined that this order, at this size, should be placed at all. and the other describes the routeHow the order travelled from wherever it was decided to the market.. The gap between the decision and the route is not a subtle difference of emphasis between two overlapping ideas. The two are answers to two different questions about the same order, and answering one of them does not begin on the other.
Because the decision and the route are separate facts about an order, neither one constrains the other, and an arrangement can pair any decision with any route. A rule can decide and a person can then type the resulting order into a screen by hand. A person can decide and a program can carry the order out through a connection at a speed no typist could match. Neither of those is a strange edge case invented for a lesson. Both of them sit inside ordinary firms today. Neither word can stand in for the other.
There is a habit of language underneath the confusion, and it is worth naming. People describe arrangements by the part of them they can see. The connection is visible, documented, switched on by somebody and named in an agreement. The decision is invisible: it lives in a rule that may be four lines long inside a sheet nobody outside the room has opened. So the visible half gets used as the name for the whole thing. The invisible half, the one the regulation of decisions is about, is never described at all.
Can a firm be doing algorithmic trading without doing API trading?
The decision has two states and the route has two states. How many arrangements does that produce?
What does a person deciding through a programmatic interface look like?
Take the first combinationThe pairing of one kind of decision with one kind of route, which is what an arrangement actually is. off the shelf, the one that catches people. A person sits and decides every order personally. He looks, he thinks, he decides to buy, and then instead of typing it into a screen he presses a key in a small program of his own that hands the order straight to the connection. His reason has nothing to do with rules or strategies. The screen is slow, his hands are slower, and the program saves him the typing.
The route was automated and the decision was not, so a person deciding each order and sending it through a programmatic interface is API trading and is not algorithmic trading. Ask him what decided the order and the answer is a man in a chair. Ask him how it travelled and the answer is a connection. Both answers are true at once, and only one of them has anything to do with the word algorithmic.
The everyday version is a shopkeeper who buys a card machine. The decision to sell three kilos of rice to the person in front of him is his, exactly as it was when he took notes and coins. The route the payment travelled changed. Nobody would say the card machine bought the rice.
Can a system decide while the order still goes out through a screen?
Now turn it around. The reversed combination is the one people find hardest to believe until they meet one. A rule decides. The rule watches a price through the day and produces an instruction: buy this quantity now. But there is no connection into the market at that firm, so the instruction appears on a screen in the dealing room and somebody types it into an ordinary trading terminal exactly as written, without deciding anything about it.
Where a person types in an order that a rule produced without judging it, the decision was still made by the system, and the presence of a human pair of hands on a keyboard does not move it. A keyboard in the path matters less than it looks. The typist is a route, not a decision maker. Asked why that order and that size, she has no answer except that the rule said so. A connection would give the same answer if a connection could be asked.
The pause while she reads and types changes the speed, and speed is not the test. Nothing in the definition of algorithmic trading concerns how quickly the decision reaches the market. The definition concerns who or what made the decision.
What do the other two squares look like, and which one do people mean?
The third combination is a system deciding and a programmatic interface carrying, and this is what nearly everybody means by either word. The rule decides, the connection carries, and the whole arrangement runs without a person in the loop for either half. Because both halves are automated at once, either word describes it correctly, and that is precisely how the two words got welded together in ordinary speech.
The fourth is manual tradingA person deciding each order individually and entering it themselves.: a person decides and a screen carries. Anasuya Kolhapure held 1,200 shares in her account with Bhadra Securities Private Limited and decided one morning to sell 400 of them, leaving 800. She thought about it, she placed it herself, and no rule and no connection was involved at either end. Ordinary trading is one square out of four rather than the natural state of the world, and it is the only square where neither word applies.
A person decides each order personally and sends it through a programmatic interface. Which square is that?
Which regulatory questions does each of the two raise?
Separate the words and the questions separate with them. The practical payoff of the whole distinction is exactly that. Questions about the decision are questions about the rule: what it is, who wrote it, who may change it, what it does when the market does something it was never shown, and who is watching it while it runs. Questions about the route are questions about the connection: who may use it, what it accepts, what happens when it fails and what record it leaves behind.
Each half raises its own questions, and answering one half thoroughly leaves the other half entirely open. Yashodhan Pai, the compliance officer at Bhadra Securities Private Limited, sees the mistake made in a particular way. A firm audits its connections carefully, documents who has access, tests what happens when the link drops, and then believes it has dealt with the subject. It has dealt with one half. Nobody has yet asked what decides the orders that go down that link.
Everyday version: a hospital can have an immaculate ambulance service, tested weekly, with drivers trained and routes mapped. An immaculate ambulance service says nothing at all about who decided the patient needed to be moved. Both questions matter and neither answers the other.
A firm satisfies itself that its connection arrangements are in good order. What has it not yet answered?
Why did the two labels collapse into one word?
Labels get made where the cases are common, and this one is no exception. The arrangement people met first, and meet most often, is the one in the top right square: a system decides and a programmatic interface carries. When something always appears with a companion, ordinary speech gives the pair one name. Nobody says a bicycle and its wheels. The wheels are assumed.
The single word works for as long as the two halves arrive together. The word fails on exactly the cases where somebody needs it most, the arrangements carrying one half without the other. The whole confusion has that shape, and the error is hard to catch from inside for that reason. Somebody using the word loosely has been right every time so far. A speaker who has never met a case where the looseness cost anything has never been taught that the word has two halves at all.
The correction feels pedantic the first time it is made and obvious afterwards. A word that is right in nine situations out of ten does not feel like a broken word. The word feels like one that works, with an awkward exception, right up until the exception is the situation at hand.
Why did one word come to stand for both halves?
How does one tell which of the two somebody is describing?
One question does it, and it takes four seconds. What decided this order? Not what sent it, not what it ran on, not how fast it went and not what it was called in the sales brochure. What decided it. If the answer names a person choosing this order on this morning, the arrangement is not algorithmic trading, whatever the connection is. If the answer names a rule, it is, whatever the connection is.
Everybody reaches first for the connection, and the connection answers the same way in three of the four squares, so looking at the connection never works. A test that gives an identical reading for three different situations is not a weak test. It is not a test at all. AutomationAny arrangement where a system acts without a person acting each time. is present in three of the four squares, and in each of them it is doing a different job.
The claim a firm commonly makes, that it trades through an interface, describes the route and says nothing about the decision, and the honest response to it is a question rather than a conclusion.
What single question separates the two?
A client has built a spreadsheet that watches prices and sends orders. Would that client describe himself as an algorithmic trader?
What does this look like at one firm, on one morning?
Bhadra Securities Private Limited is a broker and a depository participant with 11,400 client accounts on its books. On one ordinary morning, three separate arrangements are sending orders to the market through the firm, and they land in three different squares of the drawing above.
The first is the firm's own. A rule set that Sumana Rege's team runs watches prices and produces orders, and those orders go out through a programmatic interface. A system decided and an interface carried, and both words describe it correctly. The second is a client with a spreadsheet. The sheet watches prices through the day, and when a condition is met it sends an order through the same interface the firm offers to any client who asks. A system decided and an interface carried. The arrangement sits in the same square as the firm's own, and the client has never once described it that way. The third is a different client who uses the interface because typing is slow. He decides every order himself and the connection merely carries it faster than his hands could. A person decided and an interface carried.
All three arrangements reached the market through the identical connection, and the connection distinguished none of them from the others. Put the question to Yashodhan Pai and he does not open the access list. He asks, of each of the three, what decided the order. The decision question separates them in one pass, and nothing else on the morning does.
Three arrangements at one firm used the identical connection. What does that show about the connection as a test?
The reading that goes wrong: reasoning from the connection
Here is the mistake, and it is made by careful people. Everybody at the firm is using the same connection, so it feels obvious that the same answer must apply to all of them, either that the rules bite on all of them or that they bite on none. The route gets treated as the deciding fact. It is not. The decision settles it, and the route was never in the running.
The cost of this falls hardest on the client with the spreadsheet, whose arrangement turns out to sit inside a definition he had no reason to think was about him. He was not being careless. He was using the word the way everybody around him uses it: an algorithm, in ordinary speech, is a large thing that a large firm builds with a team behind it, and his sheet has four lines in it and lives on one laptop. Nothing in the ordinary use of the word told him it was about him, and nothing about the size of a system enters the test at all.
The correction is worth making gently and early. A person in this position has not cut a corner. He has used a common word in its common sense, and then discovered that a defined term and a common word share a spelling. The useful move is to write down what decided and what carried, and read from there.
What does a firm settle before it asks which rules apply?
The first move is not to open a rulebook but to write two lines down, in this order: what decided, and what carried. Two lines, in plain words, for each arrangement in the building. Two lines are the whole exercise, and a firm that has written them honestly has done the part that actually needs judgement.
Writing the combination down before reading anything is what stops a firm from looking up the answer to a question it has not identified yet. The order of the two lines matters as well. The decision is the half people skip, so write it first. The route was going to be mentioned anyway, so write it second. Yashodhan Pai keeps that pair of lines for the firm's own strategy and for every client arrangement that comes to his attention, and the pair of lines is what he reads from when somebody asks a question about any of them.
How does a compliance officer, an analyst or an investor use any of this?
Take three readers in turn. The same distinction pays off differently for each. A compliance officer at a broker uses it as an inventory question. Not how many clients have a connection, a list anybody can print, but how many of those connections have a system on the other end of them. The two counts are different numbers and the second one is the one that matters, and at a firm with 11,400 client accounts the second count is not obtainable from the access list at all.
An analyst looking at a broking business uses it to read a disclosure properly. A firm that says a large share of its volume comes through interfaces has told the analyst about its plumbing, a real fact about its technology spending and its client mix, and has said nothing about how many decisions in that flow were made by systems. Two firms with identical route statistics can have completely different answers to the decision question, and those answers point at different sorts of exposure.
An investor who is offered a service uses the distinction as a question rather than as a conclusion, and the question is what decides the orders in my account. If the answer is a rule, the investor is in one square and there is a whole subject to ask about. If the answer is a person, the investor is in another. The everyday shape of it is familiar: knowing that a courier delivers a medicine is not the same as knowing who wrote the prescription, and no amount of detail about the courier reveals the second thing.
Where does a reader look up what applies to each half?
Requirements around algorithmic trading in India are set by the Securities and Exchange Board of India and read at sebi.gov.in, and the exchanges publish operational requirements for their own members, read at nseindia.com and bseindia.com. Conditions attaching to access arrangements are read in the same places. Which of those readings is the right one is settled by asking what decided the order, and that question comes first. A requirement looked up for the wrong half is worse than no reading at all.
References
| Source | Document | Where |
|---|---|---|
| Securities and Exchange Board of India | The requirements applying to algorithmic trading, attaching to the decision a system makes rather than to the connection an order travels down | sebi.gov.in |
| National Stock Exchange of India | The operational requirements an exchange sets for its own members around automated arrangements and access, additional to those of the regulator | nseindia.com |
| BSE Limited | The equivalent operational requirements set by the Bombay Stock Exchange (BSE) for its members, applying to whichever venue an order is actually sent to | bseindia.com |
| International Organization of Securities Commissions | The principles on automated trading and market integrity, the international frame around decisions taken by systems | iosco.org |
Bhadra Securities Private Limited, Anasuya Kolhapure, Yashodhan Pai and Sumana Rege are invented.
Educational material. Not advice on any investment, tax, budget or market position.