Register
Learn guide

Understand the Risks of Trading Automation

Automating a strategy removes some risks and adds others. The strategy still loses when the market moves against it, and the automation itself can fail: a bad rule, a stale price, a rejected order or a broker session that has ended can leave you with positions you did not plan. This guide lists the risks of algorithmic trading for a retail trader in India, with the controls that limit each one.

QuantPower Research · Updated 16 Nov 2026 · 18 min read

Two kinds of risk run through every automated strategy. Market risk is the chance that the rules lose money because prices move against them, and it is the same whether you trade by hand or by algo. Operational risk is the chance that the machinery around the rules fails: the wrong order, the right order at the wrong size, an order that never arrives, or one that arrives twice. A manual trader faces mostly the first kind. An algo trader faces both, and the second kind tends to strike fast, because software repeats a mistake as quickly as it repeats a good trade.

Algo trading risks at a glance

Diagram of the two kinds of algo trading risk, market and operational, with the main control for each (illustration)
KindExamplesMain control
Market riskThe strategy stops working; the market gaps; liquidity dries up; leverage magnifies a lossPosition sizing, defined-risk structures, testing on unseen data
Operational riskA wrong rule; a duplicate order; a broker rejection; a stale price; a session that has endedLimits on orders and positions, a daily loss limit, a kill switch, daily checks
Regulatory riskStrategies from unregistered sellers; return promises; account sharingTrading only through your own broker account under SEBI's framework
Personal riskOver-trust in a backtest; switching rules after a loss; trading too largeA written plan, a size rule, and a review routine

Automation changes the balance between these. It removes emotional exits and missed entries, and it adds failure modes that a manual trader never meets. Each deserves a plan before the first live order.

A short history of algo trading failures

The risks on this page are not theoretical. Each has caused a large, public loss.

On 6 May 2010, US stock indices fell by about 9% within minutes and recovered most of the fall soon after. A later joint report by the US Securities and Exchange Commission and the Commodity Futures Trading Commission traced the start of the "flash crash" to one large automated sell order in index futures, executed without regard to price or time, which met algorithms that withdrew from the market as prices fell.1

On 1 August 2012, Knight Capital, then one of the largest market makers in US shares, deployed new trading software to its servers. One server still carried old, unused code that the new release switched back on by mistake. In about 45 minutes the firm's systems sent millions of orders and built positions it never meant to hold, at a loss of about 440 million US dollars.2 The cause was not a bad strategy. It was a deployment that nobody checked on every machine.

India had its own lesson two months later. On 5 October 2012 a dealing error at a brokerage firm sent a large basket of sell orders into the NSE's cash market, and the Nifty fell about 15% in minutes before the exchange's circuit breakers halted trading for 15 minutes.3 Exchanges and SEBI tightened order-value checks and price bands soon after, and SEBI had already published its first broad guidelines on algorithmic trading on 30 March 2012, asking exchanges and brokers to put order-level risk checks in place.4

Exchange outages are part of the history too. On 24 February 2021 the NSE halted trading in the cash and F&O segments for several hours because of a telecom link problem, and extended the session into the evening once trading resumed.5 Traders whose positions depended on a stop that the platform was watching learned how much depends on systems they do not control.

For retail traders the most recent chapter is regulatory. SEBI's circular of 4 February 2025 on safer participation of retail investors in algorithmic trading made the broker responsible for every algo order, asked exchanges to empanel algo providers, and tied API access to a static IP registered with the broker. The framework came into force for every broker from 1 April 2026.6

Market risks

The strategy stops working

Every strategy needs certain conditions. A short straddle needs a range; a breakout strategy needs trends; a mean-reversion rule needs prices to come back. When conditions change, a strategy that worked for months can lose for weeks. A backtest shows how the rules behaved in the past, not whether they will keep behaving that way.

Controls: compare live results with the backtest every month; decide in advance the drawdown at which you will pause the strategy; and test the rules on data they were not built on, for example with walk-forward testing, before going live.

Overfitting

A strategy tuned to past data, with many settings chosen to maximise the backtest, often fails live because it learned the noise of one period. The more versions of a rule you try, the more likely the best-looking one is luck.

Controls: few settings; an out-of-sample period; and a preference for rules that work across neighbouring settings, not only at one exact value.

Gaps and fast markets

A stop cannot exit at a price the market skips. Overnight positions face the next morning's opening gap, and intraday positions face sudden moves on news. SEBI's study of individual F&O traders, published in September 2024, found that 93% of them lost money over the three years to March 2024.7 Large moves against short options account for a big share of such losses, because a short option's loss grows faster as the move continues.

Controls: defined-risk structures, such as spreads instead of naked short options, for positions held overnight; a position size that survives a gap of 3% or more; and an account-level loss control, described below.

Liquidity and slippage

A strategy that backtested on candle prices may not get those prices live. Thin option strikes, market orders and large sizes all cost more than the test assumed. A strategy that makes ₹6 a lot on average before slippage can lose money once each fill costs ₹1 or ₹2 more than the test price on several legs.

Controls: limit orders with a small buffer; liquid strikes near the money; and a slippage setting in every backtest, so the result already pays for imperfect fills. The slippage calculator estimates the cost for your own size and spread.

Leverage

F&O margins let a small account control a large position. A stop that is 2% away on the index looks small, but if the position is worth ten times the capital, a 2% move against it costs 20% of the account.

Controls: size every position by the loss at the stop, not by the margin available. For example, with ₹5,00,000 of capital and a rule that no trade may lose more than 1% of capital, the most you can lose on one trade is ₹5,000. If the stop on Nifty futures is 50 points away and a lot is 65 units, one lot risks 50 × 65 = ₹3,250, so one lot fits and two do not.

Operational risks

Wrong rules

A rule that reads "sell" where you meant "buy", a quantity in lots entered as units, or a strike picked from the wrong expiry sends real orders. The algo does exactly what you told it, not what you meant.

Controls: run every new or changed strategy in Forward test first, on live prices with test money, and watch its first trades; read the activity log line by line; and keep the first live run to the smallest size.

Duplicate and runaway orders

A strategy switched on twice, an order resent after a timeout, or a rule that fires on every tick can place many orders in seconds. Under SEBI's framework, client algos that are not registered must stay within 10 orders a second, and brokers apply their own limits, but a runaway rule can still build far more position than you intended within those limits.

Controls: a cap on orders a day; a limit on how many algos run at once; checks that stop an order being placed twice; and a kill switch that stops trading in one action.

Order rejections and half-filled spreads

Brokers reject orders for insufficient margin, a wrong quantity, a banned security, a price outside the band and other reasons. A strategy that enters one leg of a spread and has the other rejected holds a naked position it never planned.

Controls: place bought legs before sold legs, so the hedge exists first; check margin before entry; and have a rule for what happens when an entry leg is refused. Our guide to troubleshooting algo trading lists the common rejection reasons and what each one means.

Stale or wrong data

A strategy that acts on a frozen price feed, or on a single bad print, can enter at the wrong moment. Indicator rules are exposed when a candle arrives late.

Controls: act on closed candles rather than every tick; refuse to decide when prices or Greeks are missing; and treat "no answer from the broker" as "unknown", never as "no position".

Broker sessions and outages

Every broker ends API sessions daily, and some need a fresh sign-in each morning. An algo that is meant to enter at 9:20 am cannot do so if the session ended overnight and nobody signed in. Brokers and exchanges can also have outages during market hours.

Controls: sign in to each broker before the market opens; know how to exit every position by hand in your broker's own app; and keep that app logged in as a backup.

Static IP and network rules

Since 1 April 2026, orders sent through a broker's API must come from a static IP registered with that broker. An order from any other address is refused. A strategy that worked on a home connection can stop placing orders the day the rule applies to its route.

Controls: use a platform that provides a registered static IP, or arrange your own, and check that the address on the broker's whitelist matches the one your orders actually leave from.

  • Unregistered providers. Under SEBI's framework, algos offered to retail traders go through a broker and are registered with the exchange. A strategy bought from an unregistered seller has none of those checks.
  • Black-box strategies. A strategy whose rules are hidden from you counts as a black box under the framework, and its provider needs research analyst registration. If you cannot read the rules, you cannot judge their risk.
  • Return claims. Brokers and their partners may not refer to the past or expected returns of an algo. A provider that promises a monthly return is outside the rules.
  • Account sharing. Letting someone else operate your account, or operating theirs, breaches broker terms and SEBI's rules unless the operator is registered to do so.
  • Tax. Automated trading can produce hundreds of trades a year, each of which has to be reported. F&O income is normally business income, and losses carry their own set-off rules.

Our guide to SEBI rules for option trading explains the framework in full.

Personal risks

The last set of risks sits with the trader, not the software:

  • Over-trust in a backtest. A clean equity curve invites a large first position. Treat a backtest as a test of the rules, not as a forecast.
  • Changing rules after a loss. Switching a strategy off after three losing days and on again after three winning days turns a systematic plan back into a discretionary one, usually at the worst moments.
  • Running too many strategies. Ten strategies each with a sensible stop can still lose together on one bad day. The combined loss needs its own limit.
  • Not watching. "Fully automated" does not mean "unattended". Someone has to read the log, sign in to the broker and check the positions.

Worked example: one bad day, three accounts

Take three traders who each run the same Nifty short straddle with one lot (65 units) and a 30% stop on each leg, and who each have ₹3,00,000 in their account. On a day when Nifty moves 1.5% in the first hour, both legs' stops trigger in turn. With each leg sold at about ₹157, two 30% stops cost 2 × 47 × 65, about ₹6,100, and the strategy loses about ₹6,300 including charges.

  • Trader A runs one lot and has a daily loss limit of ₹15,000. The loss is about 2% of capital; the limit is not reached, and nothing else happens.
  • Trader B runs five lots because the margin allows it. The same day costs about ₹31,500, or more than 10% of capital. Two more days like it would take nearly a third of the account.
  • Trader C runs one lot, but in three copies of the strategy that were switched on by mistake over the week. The loss is about ₹19,000, and the cause is operational, not market risk: nobody checked how many algos were running.

The market was the same for all three. The size rule and the operational checks decided the result.

Worked example: sizing by the stop

Suppose your account holds ₹4,00,000 and your rule is to risk at most 1% of capital, ₹4,000, on any trade.

TradeStop distanceLoss per lot at the stopLots allowed
Nifty futures, lot 6540 points40 × 65 = ₹2,6001
Bank Nifty futures, lot 30120 points120 × 30 = ₹3,6001
Nifty option bought at ₹120, stop at ₹80₹4040 × 65 = ₹2,6001
Nifty option bought at ₹60, stop at ₹45₹1515 × 65 = ₹9754

The margin for any of these trades might allow five or ten lots. The size rule allows one to four. The position size calculator does this arithmetic for any instrument.

A risk checklist before going live

  1. The strategy has passed a backtest, a test on data it was not built on, and at least two weeks of Forward test.
  2. Position size is set by the loss at the stop, not by available margin.
  3. Bought legs are placed before sold legs.
  4. A daily loss limit covers the whole account, across every broker.
  5. There is a cap on orders a day and a limit on running algos.
  6. You know how to stop all trading in one action, both on the platform and in your broker's app.
  7. You sign in to each broker before the market opens.
  8. Your registered static IP matches the address your orders leave from.
  9. You read the day's activity log and positions after the close.
  10. You have written down the drawdown at which you will pause the strategy.

When several strategies fail together

Most traders who automate end up running more than one strategy: a morning straddle, an expiry-day iron fly, a breakout on Bank Nifty. Each has its own stop, and each looks safe on its own. The danger is that they are rarely independent. Short-premium strategies on Nifty and Bank Nifty all lose on the same day of a sharp move, because they are all short volatility on closely linked indices. A trend-following rule may profit on that day, but if it trades one lot against five lots of short premium, it will not offset much.

Three habits help:

  • Group strategies by what they need. List each one as "needs a range", "needs a trend", "needs volatility to fall" or "needs volatility to rise". If most of your capital sits in one group, one kind of day will hurt all of it.
  • Add up the worst case. For each strategy, write the loss if every stop is hit at once and the market gaps through it. Add the figures. If the total is more than you would accept on one day, cut the size or the number of strategies.
  • Put a limit above the strategies. An account-level limit, such as a daily loss limit across every broker, catches the day when many reasonable stops add up to an unreasonable loss.

A daily routine for algo traders

Most operational failures are caught by a short routine, done at the same time every day:

Before the open (about 8:45 to 9:10 am)

  1. Sign in to each broker that needs a daily sign-in, and check that each connection shows as signed in.
  2. Check which algos are switched on, live or in Forward test, and that each is meant to be.
  3. Check the margin available against the margin the day's entries need.
  4. Look for events that change the day: an expiry, a holiday next day, a results announcement, a policy meeting.

During the session

  1. Glance at the activity log after each scheduled entry, to confirm that every leg filled.
  2. Watch the account's running profit and loss against the daily loss limit, not only each strategy's.

After the close

  1. Read the day's activity log line by line: entries, exits, re-entries, refusals.
  2. Compare the fills with the prices the strategy expected, and note any slippage that is larger than the backtest assumed.
  3. Record the day's result per strategy in a journal, so the monthly review compares live results with the backtest on real numbers.

The routine takes about fifteen minutes on a normal day, and it is the difference between finding a duplicated algo on day one and finding it at the end of the month.

What to do when something goes wrong

When an algo does something you did not expect, the order of actions matters:

  1. Stop new orders first. Use the kill switch for the affected segment, or pause the algo so it makes no new entries. Do this before you start investigating.
  2. Check what is actually open. Read the positions as your broker holds them, not as the strategy believes them to be. A refused exit or a part-fill can leave a position the strategy no longer tracks.
  3. Close what you did not mean to hold. Exit unplanned positions by hand, in the app or in your broker's own app if the platform cannot be reached.
  4. Read the log. Find the first line where events diverged from the plan: a refused leg, a missing price, a second switch-on.
  5. Fix and retest. Correct the rule or the setting, run it in Forward test again, and only then switch it back on live at the smallest size.

Writing these five steps on paper next to your screen sounds old-fashioned, and it works: in a fast market, people forget the order of actions they know well.

How QuantPower handles these risks

QuantPower is trading software: it runs your rules on your own broker account and does not decide what to trade. These are the controls it puts around them, as built into the app:8

  • Testing first. Rule backtests run on up to 2,000 past candles from your broker, and option strategy backtests run minute by minute and hold back the last 30% of days as an unseen check. Every algo can be switched on in Forward test, which runs the same engine with test money.
  • Account safety above every algo. Smart Exit sets a profit target, a stop-loss and a trailing step for the whole account; on a hit it stops the algos and squares off. The daily loss limit refuses new orders once today's loss reaches it. An optional cap limits orders a day. The kill switch stops trading in the segments you choose until each segment's next session and cancels their open orders. The loss figure adds up every connected broker.
  • Order safety. Bought legs go first and sold legs only after the buys fill. If an entry is refused, the algo closes the legs already in and stops, and says why. If prices or Greeks are missing, the algo waits and never guesses a strike. Each order carries a key so a retried request never places twice.
  • Limits. Up to 20 algos run at a time, each with up to 6 legs.
  • Sessions. A live algo cannot be switched on while its broker is signed out for an entry due now, and the card warns before a later entry.
  • Static IP. Every account gets 2 static IPs free, a primary and a backup. Orders leave only from them; if the proxy is down, orders are held, never sent from another address.
  • Visibility. Every entry, exit and refusal is one line in the algo's activity log.

Your phone or computer does not have to stay on while the algos trade.

Start with 500 free tokens

Connect your broker, build a strategy and test it on paper before a single live order goes out.

Where algo risk control is going

Three changes are under way. Regulation now puts the broker in charge of every algo order, so controls that were optional for a retail trader, such as order-rate limits and a registered address, are now built into the route itself. Platforms are moving controls up from each strategy to the whole account, because the losses that hurt most come from several strategies failing together. And AI tools that write strategies from a sentence make it easy to build many rules quickly, which makes testing on unseen data and a hard daily loss limit more important, not less.

Frequently asked questions

Is algo trading riskier than manual trading?

The market risk of the same strategy is the same. Automation removes emotional errors and missed exits but adds operational risks such as wrong rules, duplicate orders and broker sessions that have ended. With the controls above, many traders find automated execution more consistent than their own.

What is the biggest risk for a new algo trader?

An untested rule or a wrong setting that places real orders. Running every new or changed strategy in Forward test, at the smallest size, is the simplest defence.

Does my computer need to stay on while an algo runs?

Not on QuantPower. The algos keep running with your computer off and place orders through your broker's official API from your registered static IPs. You still need to sign in to your broker each day where its rules require it.

Can an algo lose more than my account balance?

In theory yes, with short options or futures through a large gap. Defined-risk structures, sizing by the stop and an account-level loss control keep the loss within a planned amount.

How do I stop all my algos at once?

On QuantPower, the kill switch stops trading in the segments you choose until each segment's next session, and Exit all stops the running algos first and then squares off. If the platform cannot be reached, close the positions in your broker's own app.

How much capital should I risk per trade in an algo?

Many traders cap the loss on a single trade at 1% to 2% of capital, and the loss on a single day at 3% to 5%. The right figures depend on your capital and how many strategies run together; the point is to decide them before the first trade, not after a loss.

Should I run several strategies at once?

Several strategies that need different conditions can smooth results, but several that need the same conditions add risk. Check what each one needs, add up the worst case, and set an account-level limit above them all.

References

  1. US SEC and CFTC, "Findings Regarding the Market Events of May 6, 2010", 30 September 2010 (sec.gov)
  2. US SEC, order in the matter of Knight Capital Americas LLC, 16 October 2013 (sec.gov)
  3. NSE, press release on the trading halt of 5 October 2012 (nseindia.com)
  4. SEBI, broad guidelines on algorithmic trading, circular CIR/MRD/DP/09/2012, 30 March 2012 (sebi.gov.in)
  5. NSE, press releases on the trading halt and extended session of 24 February 2021 (nseindia.com)
  6. SEBI, circular on safer participation of retail investors in algorithmic trading, 4 February 2025, in force for every broker from 1 April 2026 (sebi.gov.in)
  7. SEBI, "Analysis of profit and loss of individual traders dealing in equity F&O segment", 23 September 2024 (sebi.gov.in)
  8. QuantPower: risk controls, order checks and limits, from the app's code, checked 12 November 2026
Trading involves substantial risk. This page is for education and is not investment advice.