Guide · Costs and taxes · Updated 2026-09-18

A trading loss is one of three things, the set-off between them runs one way, and the deadline destroys it

Two errors dominate this subject. The first is stating the set-off symmetrically, which reads tidy and is false in one direction. The second is treating the filing date as an administrative matter with an interest cost attached, when for a loss year it is the only thing between a usable carry-forward and nothing at all. There is now a third: until this year a section number was a section number, and from 1 April 2026 it is not.

Two statutes, and the year you are standing in decides which one is yours

The Income-tax Act 2025 came into force on 1 April 2026 and replaced the Income-tax Act 1961. In this area it is a renumbering and a restructuring rather than a change of policy, but the renumbering is near total across its 536 sections. It also retires a duality every Indian filer has carried for six decades: the 1961 Act separated the previous year, in which income was earned, from the assessment year, in which it was taxed, and the 2025 Act collapses both into a single tax year running 1 April to 31 March, which is the year the income was earned in.

That matters here more than elsewhere, because carry-forward windows are counted in years and the years now have different names. A speculative loss was carried four assessment years following the assessment year it was first computed in. It is now carried four tax years following the tax year it was first computed in. The span is the same length; the label is not, and a working paper that mixes the two counts the window twice or not at all.

The 2025 Act carries transitional provisions preserving losses brought forward from years beginning before 1 April 2026, so a loss properly determined under the old statute does not evaporate on the changeover. It continues under a new section number with the remaining years counted in the new label. The loss survives. The citation does not.

Character first, and the settlement mechanism decides it

Before any set-off question can be asked, each rupee of loss has to be given a character, and the statute assigns it rather than the trader. The test is how the contract was settled. A speculative transaction is one in which a contract for the purchase or sale of a commodity, including stocks and shares, is settled otherwise than by actual delivery or transfer. That is section 43(5) of the 1961 Act, and section 66(31) of the 2025 Act.

An intraday equity position squared off before the session ends is settled by difference, never reaches the demat account, and is inside that definition. Exchange-traded derivatives are also settled without delivery, and would sit inside it but for an express carve-out for an eligible derivative transaction on a recognised stock exchange, which survives the change of statute and sits within section 66 alongside the definition. Delivery-based equity is outside it, and falls under capital gains unless volume, frequency and intent make it a business on its own facts.

One trap sits outside that table and catches traders who operate through a company. A deeming fiction treats the purchase and sale of shares of other companies as a speculation business to that extent, regardless of whether delivery was taken, subject to exceptions including where the principal business is trading in shares, banking or the granting of loans and advances. It is the Explanation to section 73 of the 1961 Act, carried into section 113 of the 2025 Act. An individual trader never meets it. A trader who incorporated for other reasons meets it on day one.

The set-off is one-way, and almost every page writes it both ways

Search the question and the answer that comes back is almost always some version of "speculative and non-speculative losses cannot be set off against each other". That sentence is half right, and the half it gets wrong is the half a losing derivatives book depends on.

The mechanism sits in two provisions doing different jobs. The first permits intra-head set-off: where the net result from any source under a head is a loss, it may be set off against income from any other source under the same head. Section 70 of the 1961 Act, section 108 of the 2025 Act. Speculative and non-speculative business are two sources under one head, so this reaches across them.

The second restricts one of those flows and only one. A loss computed in respect of a speculation business shall be set off only against the profits and gains of another speculation business. Section 73(1) of the 1961 Act, section 113 of the 2025 Act. It is written as a restriction on the speculative loss and says nothing about a non-speculative loss, because it is not addressed to one.

So the two directions are governed by different amounts of law. One way, a permitting provision meets a restricting provision and the restriction wins. The other way, the permitting provision meets nothing. A derivatives loss may absorb an intraday profit in full. An intraday loss may not touch a derivatives profit at all.

This was checked in both statutes rather than assumed. The restricting language in section 113 of the 2025 Act is one-directional in the same way section 73(1) of the 1961 Act is, and no provision was located in the 2025 Act barring a non-speculative loss from meeting speculative income. Where a position turns on a section number rather than on the substance, the Income Tax Department publishes a utility reading the 1961 and 2025 provisions side by side, and that is the check to run before using any mapping, including the one below.

The consequence is a valuation, not a technicality. A non-speculative loss is worth strictly more than a speculative loss of the same size, because it reaches more income in its own year and more years afterwards. Two traders who lost the same amount on the same day did not lose the same thing.

The same two figures, reversed, and the outcome is not symmetric

Take one year and run it twice with the signs swapped. Illustrative figures throughout.

The same six figures, arranged two ways. In Case A, three lakh of income is taxed now and 1,80,000 of loss is parked against a future that may never contain speculative income. In Case B the whole loss is used inside the year and nothing is assessed. The difference is not the trading, it is which book the loss landed in. It is also why the instinct to net the two books fails in exactly one case: netting Case A gives business income of 60,000 where the correct figure is 2,40,000, while netting Case B lands on the right answer by accident, which is worse than being wrong because it confirms the method that fails on the other case.

Intra-head first, then across heads, in that order

Set-off runs in a fixed sequence, and running it out of order wastes relief that was available.

One. Within the head. Losses from one source are set off against income from another source under the same head, subject to the speculation restriction above. Section 70 of the 1961 Act, section 108 of the 2025 Act.

Two. Across heads. What remains meets income under other heads, under section 71 of the 1961 Act and section 109 of the 2025 Act. Two bars apply here. A business loss may not be set off against salary, expressly, under section 71(2A) and section 109 respectively. And a house property loss is capped at 2,00,000 of inter-head set-off in a year, with the excess carried forward.

Three. What is left is carried forward under the section matching its character, with the window and the narrowed scope that section imposes.

Capital losses never leave their head at any step. A short-term capital loss meets both short-term and long-term capital gains; a long-term capital loss meets long-term gains only. Neither reaches business income, in its own year or afterwards.

Four against eight, and what a carried loss gives up

The windows sit in the sections themselves, and the table below states each. What the diagram adds is the half left out of nearly every summary: a carried-forward loss is not the same instrument as the loss that arose. In its own year a non-speculative business loss reaches interest, rent and capital gains, everything but salary. Once carried, it reaches business income and nothing else. Carrying preserves the amount and narrows its reach.

For a speculative loss the narrowing is worse, because it was already confined to speculative income in its own year, so carrying it gives it four more attempts at the same narrow target. A trader who stops trading intraday after a bad year holds a speculative carry-forward with nothing it is permitted to meet, and it runs out its four years against an empty column.

One point runs the other way. A non-speculative carry-forward is not defeated by the business being discontinued; it can still be set off against the profits of a business carried on later. The condition is on the income it meets, not on the survival of the business that produced it.

The cliff at the due date, which is where most losses actually die

Everything above assumes the loss exists. Whether it exists is decided by a single date, under a provision most traders have never read.

Section 80 of the 1961 Act provides that no loss which has not been determined in pursuance of a return filed within the time allowed shall be carried forward and set off under the business, speculation, capital gains or race-horse provisions. Section 121 of the 2025 Act carries the same condition, tied to a return filed under section 263(1) rather than section 139(1).

Read what that does and does not do. It does not reduce the loss, charge anything, invalidate the return or stop the department computing tax on it. It removes the loss from the carry-forward provisions entirely, so next year there is nothing to bring forward. One day late converts an eight-year asset into nothing.

The due date under the 1961 Act is 31 July following the financial year where no audit is required, and 31 October where accounts must be audited. Those dates are extended administratively in some years and not others, so the operative date has to be confirmed rather than assumed. The section fixes the consequence, not the date.

The two survivors explain why so many believe the rule is softer than it is. House property loss and unabsorbed depreciation sit outside the forfeiture provision and carry forward on a belated return. A salaried trader who files late keeps the housing carry-forward, loses the trading one, and reasons from the first that the second must be safe. An updated return does not repair it: that facility exists to increase income, not to create or increase a carried loss.

The behaviour is what makes this expensive. A profitable year creates a liability and a liability creates urgency. A loss year creates neither. The filer sees nothing to pay, treats the deadline as soft, and learns the following year that the loss was never determined. The provision is aimed precisely at the year in which nobody feels any pressure to meet it.

Old section, new section, and the year each one governs

Every provision cited on this page, with both numbers and the years each governs. The 2025 Act numbers were corroborated across independent professional sources, and the Income Tax Department's side-by-side utility is the authority to check any mapping against.

Two cautions on it. The carve-out for exchange-traded derivatives was corroborated as sitting within section 66 alongside the definition, but the precise sub-clause should be confirmed against the official utility before being cited in a working paper. And the audit and presumptive rows are listed for orientation only: section 63 of the 2025 Act is reported to have altered the low-profit audit trigger rather than merely renumbering section 44AB, which is a separate question and outside this page.

Where the carry-forward actually dies

Filing the loss year late. The largest destroyer and the only total one. Nothing about a loss year creates urgency, which is exactly why the provision is written as it is.

Netting the two books before classifying them. Right in one direction, wrong in the other, which is why the habit survives. Classify, then set off, then carry.

Claiming a trading loss against salary. Barred expressly in both statutes. A salaried trader sees a large salary figure and a large loss figure and assumes they meet. They do not, in that year or any later one.

Treating a carried loss as though it were still fresh, or letting a speculative window lapse. Inter-head reach is lost on the way into the carry-forward, so a business loss brought forward from three years ago cannot meet this year's interest income though it could have in its own year. And four years is short against a target as narrow as speculative income.

Losing a firm's loss on a change in its constitution. Where a partner retires or dies, the firm cannot carry forward the proportion of the loss attributable to that partner beyond that partner's share of profit for the year. Section 78 of the 1961 Act, section 119 of the 2025 Act. A trading partnership that changes hands mid-year can lose part of a carry-forward with nobody in it having made a decision about tax.

Citing the right rule under the wrong statute. New this year, and it will persist, because the professional writing that ranks well was written before 1 April 2026 and does not say which statute it describes.

What the character is actually for

The three characters exist because the legislature took a view about which activities may shelter which income. Whether that view is right is not a question a filer gets to answer. What a filer gets to do is know which character each rupee of activity is producing, while the activity is still being chosen. Character is fixed the moment the order is placed, by the way the contract will settle, and nothing done at filing changes it.

The asymmetry, the four-year window against the eight, and the cliff at the due date all follow from a decision made months earlier by someone thinking about something else. A record kept during the year, split by character as the trades happen and treating the filing date as a hard constraint, answers every question on this page before any of them becomes urgent. A record reconstructed in the week before the deadline answers none of them.