Automation & Trading · October 14, 2026 · 14 min read · By CryptoTaxCalc Team

My Bot Made 12,000 Trades Last Year. The Tax Return Was the Hard Part.

I ran a grid bot on a liquid pair through a quiet quarter — a few hundred dollars of range, low fees, no news. It filled 12,000 times between January and March and turned a small book into a mountain of rows. The trading was the easy part. The return took a week, one rejected CSV from my accountant, and a hard question about whether I was still an investor at all. Here is what automation actually does to a return, in seven countries.

Image: Illustrative purposes. Rules cited from IRC §1091/§475/§1221, HMRC trading versus investment guidance, §15/§20/§23 EStG, CRA business income guidance, ITAA 1997 Divisions 6/104, 所得税法 and CBDT s.194S/§115BBH.

Here is the thing nobody tells you when you switch a bot on: the software does not think about character. It sees a spread and takes it. Twelve thousand times. Somewhere around fill 4,000, the tax question stopped being "what did I make" and became "what am I" — because a person who fills 12,000 orders a quarter with a profit motive, logged timestamps and a written strategy looks less like someone holding an asset and more like someone carrying on a business. That distinction, not the arithmetic, is what decides your rate and your losses in every country on this list.

The quarter my bot ran, in numbers

I ran the bot on Binance against USDT from January to March on a pair with tight spreads and decent depth. The range was narrow — about 6% peak to trough for the quarter — so the bot did what grid bots do: it bought small, sold small, and reprinted a fraction of a percent each time. Gross it looked respectable. Net, after taker fees and one bad tick that got filled six times in a row, it was modest. The number that mattered was the row count: 12,000 fills across roughly 4,100 round trips on the same asset, plus the funding and fee lines the exchange tacks on separately.

My accountant's first response was to ask for the exchange's annual CSV. My second email attached it. Her third email rejected it: one row per trade, but no running cost basis, fees in a separate column she could not allocate to lots, and internal transfers between my spot and bot sub-accounts showing up as unknown movements rather than as the same wallet. That rejection is the real story — automation is a data problem wearing a trading costume.

Three kinds of automation, three tax profiles

People say "bot" as if it were one thing. It is at least three, and they do not share a tax profile.

1. Your own bot, your own account. Every fill is your disposal. Full stop. The bot adds nothing legally — it has no separate personality, it does not trade "for the platform", and the fact that a machine pulled the trigger does not make the gain passive. Imagine instead that you had clicked 12,000 times by hand; that is how the tax authority sees it.
2. Copy trading as a follower. You subscribe to a lead trader and your account mirrors their orders. The fills are still yours and are still disposals in your jurisdiction, sized to your allocation. The platform's marketing — "passive", "hands-off", "let the pros trade" — has no effect on tax character. Passive in the app store, active on the return.
3. Being the lead trader. Now you are on the other side. You trade your own book and you receive a share of your followers' profits as a performance fee. That fee is not a capital gain. It is income for services you provided, it is a separate tax stream from your own trading gains, and it usually drags a VAT, GST or consumption-tax question along with it, because software-and-services revenue is precisely what those taxes target.

The volume problem compounds all three. A market-making or grid bot on a liquid pair can generate thousands of disposals per month, and each one needs a timestamp, a quantity, a price and a fee — then a basis computed in the correct order. Ordering matters more than most people admit: FIFO and average cost produce materially different gains on the same trade history, and in most of these countries you cannot pick a method per year to suit the answer you want. The basis rules for some of these positions, particularly derivatives and perpetuals, are covered in the futures, perpetuals and margin guide.

The seven-country table

Read this as the shape of the problem, not as advice. Whether you are a business or an investor is a facts-and-circumstances call, but the table below shows how much turns on it.

Business or investment: the question that changes your rate

If you take one thing from this article, take this: the character question is worth more money than any cost-basis method. An investor pays a capital-gains rate on a net figure, gets an annual exemption or allowance, and — the sting — may only use losses against capital gains. A business pays income tax at marginal or corporate rates on trading profit, gets to deduct expenses, and can generally use losses far more freely, but takes on registration, social-contribution and compliance overheads.

Automation pushes you toward the business end of that spectrum, involuntarily and by accumulation: frequency, organisation, a written strategy, and a profit motive read as badges of trade. My own 12,000 fills in a quarter were, on the face of it, a far stronger badge than anything most retail investors ever generate. The full analysis of when an active crypto account becomes a trade — with the thresholds and the tests each authority actually uses — is in the trader-status guide, and the day-trading patterns that trigger it fastest are in the day-trading tax guide. Read those two alongside this piece before finalising a bot's strategy, because the character call is made on the pattern, not the annual total.

The record-keeping rebuild

Month-end aggregation fails for a structural reason: it hides the ordering of lots. If your bot bought and sold the same asset forty times in a month, the month-end net hides which lots were sold, and if the jurisdiction requires FIFO or a pool, the sequence is the whole computation. I rebuilt my ledger three times before it held. What finally worked was this:

• API pulls, not CSV uploads. Pull the raw trade history from the exchange API and store it as received — one row per fill, with the exchange's own identifiers, so nothing is de-duplicated before you can see it.
• One normalised transaction table. Every row gets the same shape: timestamp with timezone, asset, quantity, price, fee in the fee currency, and a direction. Transfers between your own wallets are tagged as transfers, not disposals — the mistake the rejected CSV made.
• Month-by-month reconciliation. At each month end I reconciled the running asset balance in my table against the exchange's reported balance. Any drift was almost always a fee paid in the quote asset or a funding charge on a perpetual, and once found, it was fixable. Drift caught in the same month costs minutes; drift caught in April costs days.
• Treat on-chain data as a separate source. Exchange CSVs and on-chain history disagree by design: an exchange CSV sees your balance with the venue, while on-chain sees the wallet, so a withdrawal to self-custody appears as a disposal to one tool and as a transfer to the other. If you moved coins off the exchange to run anything, you have two ledgers, and they must be reconciled against each other.

The practical system I now use — the exact column layout, the reconciliation ritual and the monthly checks — is written up in the record-keeping spreadsheet system guide. It is not glamorous, and that is the point: the spreadsheet is what keeps the bots honest.

Copy trading specifics

As a follower, keep the disposal discipline of situation two above: the fills land in your account, mirror your allocation, and are taxed as your disposals. Two extra wrinkles bite. First, a copy-trading account often churns far more than you would have, because you inherited someone else's frequency — so you can pick up a badge-of-trade pattern without ever deciding to trade that way. Second, some platforms report closed positions as realised P&L in a summary that is not a tax computation; it can ignore fees, funding and the opening lot, so treat it as a hint, not a figure.

As a lead trader, split the ledger in two from day one. Your own trading gains or losses are one stream. Your performance fee is another: it is consideration for a service, it is taxed as ordinary or business income at your marginal rate, and it is the line that raises the indirect-tax question. In the UK, business services can push you toward VAT registration once turnover crosses the threshold; in Germany, a service supplied across borders can trigger reverse-charge or trade-tax analysis; in Australia and Canada a GST/HST question follows the same logic; Japan and India have their own consumption-tax and GST versions. It is just the ordinary tax on selling a service, arriving because your automation is now a small business with customers. Keep platform fee statements as evidence of the service revenue — exactly the document a VAT officer will ask for.

US: wash sales on borrowed time, and the §475 door

The American surprise is that crypto is not a "security" for IRC §1091, so the wash sale rule technically does not reach coins today. A bot that sells at a loss and rebuys the same asset within 30 days is therefore not disallowed — for now. That is a rule with a shelf life. H.R. 10357 and similar proposals would extend wash sale treatment to digital assets, and if that passes with an early effective date, exactly the behaviour your bot performs all year becomes a deferred loss you never get to claim, with no basis addition to soften it. I cannot say it will pass, and neither can anyone honest; what I can say is that a bot whose economics depend on that gap is running on a rule that may not survive, and there is a real chance the IRS argues the current position differently for a pattern this aggressive.

On the other side of the door is IRC §475. If your activity is frequent, regular and substantial enough to be a trade or business, you can elect mark-to-market treatment, which marks every position to market at year end, converts the result to ordinary income, and — the point that actually changes the maths — makes losses fully deductible against ordinary income rather than trapped behind the $3,000 annual capital-loss cap with indefinite carry-forward. The election requires timely filings, applies to securities held in connection with your trading business, and is largely a one-way door once made; you do not get to undo it casually. Absent it, every one of my 12,000 round trips is short-term, and the net loss can offset just $3,000 of ordinary income per year. If you are harvesting deliberately around year end, the mechanics are in the loss-harvesting calculator, and the current state of §1091 for coins is covered in the wash sale rule guide. The baseline US rates are on the US page.

UK: HMRC's badges and the 30-day trap

HMRC's trading-versus-investing badges matter far more to bot users than to buy-and-hold investors, and 12,000 trades with a profit motive and a high degree of organisation sit awkwardly close to "trade". That matters because a trade is taxed as income at marginal rates, not as a capital gain — you lose the £3,000 annual exempt amount, you lose the CGT rate, and you gain the trading-loss rules that are actually more useful against other income. The mechanical pain underneath it is the 30-day "bed and breakfast" rule under TCGA s.105/106: a bot that rebuys the same coin inside 30 days forces same-day and 30-day matching before anything reaches the S104 pool, and with thousands of fills the matching order becomes a genuine computation rather than a footnote. Everything outside that window pools at average cost through the S104 pool, which is why the ordering of your transactions is not optional detail. And if you are the lead trader earning profit share, or even pushing meaningful platform-fee volume as a business service, the VAT registration question arrives with the income.

Germany: from §23 to gewerbliche Einkünfte

A high-frequency algorithm can tip you out of §23 private sales and into gewerbliche Einkünfte — business income — which is a fundamentally different regime: income tax plus, possibly, trade tax, and every rebalance is a taxable event rather than a private disposal you might otherwise net. Staying in private sales has its own trap: the €1,000 Freigrenze is a cliff, not an allowance, so crossing it by one euro makes the entire gain taxable, and losses from private disposals can only be netted against other private disposal gains, never against salary. One genuinely unsettled area is the cap on offsetting losses from derivatives under §20(6) EStG, which has been litigated and reformed more than once; I am deliberately not going to give you a number, because the right answer depends on the year in question and I would be guessing past recent changes. Verify it against the current BMF guidance for your year, and treat any blog that states it flatly as out of date. Bot-triggered swaps also keep reopening the one-year holding period question, because a bot that round-trips a position is not holding anything for a year.

Canada: frequency turns capital into business

The CRA looks at frequency, expertise, time spent and profit motive, and a very active bot pattern can land you in business income rather than capital gains. The upside is real: business income removes the 50% inclusion rate — gains are fully taxed but losses are fully deductible — and losses stop being capital losses restricted to capital gains. The cost is that an individual carrying on business can owe CPP contributions on the income, which buy-and-hold investors never see. Then there is the superficial loss rule in ITA s.54/40(2)(g)(i): a 30-day rebuy of the same or an identical property can permanently deny the loss on a capital-property disposition, which is precisely the bot pattern. The rule bites capital dispositions, not business income, so the character question decides whether your bot's 30-day rebuys are a permanent loss or merely a deductible one — an honest illustration of how much rides on one determination.

Australia and Japan: CGT events and the 移動平均法

Australia's starting question is the ATO's "are you a trader?" style analysis, which decides whether your gains are ordinary income or capital gains. Run thousands of trades on a CGT basis and every single one is a CGT event, with losses only offsettable against capital gains under the CGT rules. The special feature for bot users is data reach: the ATO's crypto data-matching programme pulls exchange and platform data, including API-reported activity, so the volume is visible in a way a low-frequency, self-custodied history never was. Japan's fork is 事業所得 versus 雑所得 — business income versus miscellaneous income — which changes both the rates and whether you can deduct expenses at all; business income allows expense deductions and a loss carry-forward, 雑所得 allows neither in any meaningful way. On top of that, the default basis method for individuals is the 移動平均法, the moving-average method, which makes back-computing thousands of disposals genuinely hard if you did not record them as they happened. If there is a single argument for pulling your exchange API into a clean table monthly, Japan's moving average is it.

India: the blunt worst case

I am going to say this without softening it, because softening it has cost people real money. In India, every transfer of a virtual digital asset — including one leg of a bot's round trip — is taxable at 30% plus cess. There is no deduction for exchange fees, no deduction for gas, and no deduction for cost of acquisition beyond the narrow statutory basis. Losses from one VDA cannot be set off against gains from another, and they cannot be set off against any other income; they simply sit there. The consequence for automation is brutal and perhaps counter-intuitive: a bot that churns to a near-flat outcome can still produce a tax bill larger than its entire profit, because every profitable round trip is taxed at 30% while every losing round trip is stranded. The 1% TDS under section 194S is collected by exchanges and platforms on transfers, so it is a cash-flow event rather than your final liability. And there is no relief for algorithmic traders — the regime does not recognise the category. If you run a high-frequency bot from India, model the tax before the profit, not after. The flat regime is set out on the India page.

What I would do differently

Three changes, in order of how much they would have saved me.

1. Switch the wash-sale-adjacent behaviour off before year end. A bot that rebuys the same asset within 30 days is banking on a US rule that is actively being legislated away. Turning that leg off in December, and holding the position rather than round-tripping it, is cheap insurance against a rule that could change the character of the whole year.
2. Run the tax estimate monthly, not in April. The biggest surprise of the year was the volume, not the rate. Recomputing the gain monthly — using a consistent basis method — turns a January-to-March shock into twelve small, adjustable numbers, and surfaces the ordering problems while the data is still fresh. The calculators are enough for the estimate; you do not need a full filing to know roughly where you stand.
3. Never restart a bot on 29 December. It sounds like a joke until you see what a fresh bot does to the last two days of a tax year: a burst of disposals with fresh lots, in the window where the 30-day matching rules and the year-end cut are both live. If a bot needs a restart, do it on a date that is boring for the tax calendar, and pause it over the boundary.

My net result was small, after all of that — the honest ending of most bot stories. The bot made a modest amount, the record-keeping cost real time, and the character question made everything downstream worth re-checking. If you take one habit away, make it the monthly reconciliation: the machines are patient, the return is not.

General information, not tax advice. Whether your automated activity is investment, trading or business is a facts-and-circumstances determination that varies by year and by authority; the US §1091 wash-sale position for coins rests on current law and active legislative proposals; Germany's §20(6) derivatives-loss cap has changed repeatedly and must be verified for the specific year; and India's 30% regime has no loss-relief pathway. Confirm your own position with a qualified adviser in your jurisdiction before acting.

FAQ