SEBI Algo Trading Rules Are Live. Your Bot Still Can't Fix Your Worst Habit.

You built the algo. Backtested it across three years of Nifty data. The Sharpe ratio looked solid. The drawdown was manageable. You connected it to Kite Connect, set the static IP, registered the Algo-ID with NSE.

Then March 2026 happened. Brent crossed $115. The Sensex dropped 11.5% in a single month. FIIs pulled ₹1.14 lakh crore. Your algo, running exactly as designed, took a position on the long side during a relief rally.

You turned it off.

Not because the strategy was wrong. Not because the code had a bug. Because watching your algorithm lose ₹47,000 in a single session while you sat there doing nothing felt worse than losing ₹47,000 by pressing the buttons yourself. At least when you lose manually, you feel like you tried.

That impulse, the need to intervene when the system is doing exactly what you told it to do, is the behavioral pattern SEBI cannot regulate. And it is the reason most retail algo traders in India lose money despite having "removed emotions from trading."

What exactly changed on April 1, 2026?

SEBI's algo trading circular from February 2025 hit full compliance on April 1. Here is what is now mandatory for every algorithmic order on Indian exchanges.

Algo-ID registration. Every algorithm must carry a unique identifier issued by the exchange through your broker. NSE and BSE now track which algo generated which order. If your strategy is not registered and tagged, it cannot legally trade.

Broker accountability. Your broker owns full legal responsibility for every algo running through their platform. Zerodha, Angel One, 5paisa, Dhan: they all must audit, approve, and monitor every algorithm connected to their APIs. The era of connecting random GitHub scripts to Kite Connect with no oversight is over.

Static IP and 2FA. All API access must use a fixed IP address registered with the broker. OAuth-based two-factor authentication is mandatory. Every session must auto-logout before the next market pre-open.

Black box disclosure. If you subscribe to a proprietary algo where you cannot see the logic, the provider must now hold a SEBI Research Analyst license and periodically disclose performance metrics. No more Telegram channels selling "guaranteed 5% weekly returns" through opaque bots.

10 orders per second threshold. If your algo places 10 or more orders per second on any exchange, it must be formally registered with the exchange. Below that, you comply with the security requirements but skip the registration step.

These rules are overdue. Unregistered algos running on retail accounts contributed to flash crashes, spoofing incidents, and a market structure where 93% of individual F&O traders lost money between FY22 and FY24.

But here is what the rules do not fix.

Why does your algo lose money if it has no emotions?

The pitch for algorithmic trading is seductive. Remove the human. Remove the fear, the greed, the revenge trades, the FOMO entries. Let cold logic execute. If 93% of manual traders lose money because of behavioral mistakes, surely automating those trades solves the problem.

It does not. Because the human is still in the loop. Not in the execution, but in every decision that surrounds it.

You choose which algo to run. That choice is subject to recency bias. You pick the strategy that performed best in the last 3 months of backtesting. You ignore that the market regime that produced those returns may not repeat.

You choose when to start and stop it. You launch after a good backtest (confirmation bias). You kill it after 2 weeks of drawdown (loss aversion). The strategy needed 8 weeks to recover, but you gave it 14 days.

You choose the parameters. You over-fit the moving average period, the stop-loss percentage, the position size, all calibrated to make the backtest look perfect. This is not strategy development. This is curve fitting dressed up as discipline.

A study by Barber and Odean at UC Berkeley found that the more actively investors traded, the worse their returns, regardless of the method of execution. The mechanism of placing the order is not the variable that determines profitability. The quality of the decision, and your willingness to stick with it, is.

What is the override problem and how much does it cost?

Here is a pattern PortoAI's behavioral fingerprint system detects in algo traders who connect their Zerodha accounts.

Week 1-2: Algorithm runs. Takes positions per the rules. Small profits, small losses. Trader watches but does not intervene.

Week 3: A losing streak. Three consecutive trades hit stop-loss. The algo is down 4.2% for the month. The trader starts checking the positions mid-session instead of at day-end.

Week 4: A single large loss. ₹83,000 gone on a Bank Nifty trade that gapped through the stop. The trader logs into Kite, manually closes the next trade before the algo's exit signal triggers. "Just this once."

Week 5: The trader turns off the algo for "a few days to observe." The algo's next three signals would have been winners. The trader, now trading manually "until the market stabilizes," revenge trades on expiry day and loses ₹1.2 lakh.

This is not a hypothetical. Variations of this sequence repeat across thousands of retail algo traders every month. The algo did not fail. The human overrode the algo at exactly the point where discipline mattered most: the drawdown.

PortoAI flags this pattern as "algo override behavior." When your trading frequency suddenly spikes after a period of low, consistent activity, it means you switched from automated to manual. That switch, during a drawdown, is one of the most expensive behavioral patterns in retail trading. Our overtrading detection catches the spike before you dig the hole deeper.

Is over-optimization making your backtest lie to you?

There is a specific type of intellectual dishonesty that algo traders practice without realizing it.

You have 5 years of Nifty minute-level data. You test a mean-reversion strategy with a 14-period RSI and a 2% stop-loss. It returns 18% annually. Not bad, but not exciting. You try a 12-period RSI with a 1.7% stop-loss. Now it returns 23%. You try 11-period RSI, 1.5% stop-loss, and add a volume filter. The backtest shows 31% returns.

You have not found a better strategy. You have found the specific combination of parameters that happens to fit the noise in your historical data. This is called over-optimization, and it is the most common reason retail algos fail in live trading.

The backtest shows what would have happened if you had run exactly those parameters on exactly that data. It says nothing about what will happen when market conditions change. And market conditions always change. The Nifty that traded in a range between 17,000 and 19,000 for most of 2023 is not the Nifty that crashed from 26,000 to 22,000 between September 2024 and March 2026.

Over-optimization is a behavioral bias, not a technical error. It is confirmation bias applied to code. You keep tweaking until the numbers confirm what you want to believe: that your system works.

A useful test: if your strategy has more than 5 adjustable parameters, you are probably fitting noise. If removing any single parameter improves the backtest by less than 3%, that parameter is adding complexity without edge.

What do the new rules actually mean for someone trading below 10 orders per second?

Most retail algo traders on Zerodha Kite Connect, Dhan, or Angel One APIs trade well below the 10 OPS threshold. You are placing maybe 5-20 orders per day, not 10 per second. So the Algo-ID registration with the exchange does not apply to you.

But the operational requirements do.

Static IP. If you run your algo from home on a residential broadband connection with a dynamic IP, you need a solution. Options: a cloud server with a static IP (AWS, DigitalOcean, Google Cloud), a VPN service with a dedicated IP, or requesting a static IP from your ISP. This adds ₹500-2,000 per month in costs.

Daily auto-logout. Your API session must terminate before the next pre-open. If your algo runs on a cron job that starts at 9:00 AM, make sure it authenticates fresh each morning. Persistent sessions are not compliant.

2FA on every login. The OAuth flow requires interactive authentication. This means your fully automated "fire and forget" setup now needs a manual step each morning, or a TOTP integration if your broker supports it.

Broker due diligence. Your broker may now require you to document your strategy, even if you are below the 10 OPS threshold. Zerodha's updated API terms require disclosure of whether you are running automated strategies.

For the casual retail algo trader, the compliance burden is real but manageable. The cost is ₹1,000-3,000 per month in infrastructure, plus 5-10 minutes of daily authentication overhead.

The question you should ask is not "how do I comply?" but "does my algo actually make enough money to justify these costs?" If your algo's monthly edge is ₹5,000 and your compliance costs are ₹2,500, your effective edge just halved. Many retail algos that looked marginally profitable on a backtest will not survive the new cost structure.

Should you build an algo or fix your behavior first?

Here is the uncomfortable question.

If you are losing money trading manually because of revenge trading after losses, FOMO entries on momentum stocks, and overtrading on expiry days, building an algo does not solve the problem. It relocates it.

Instead of revenge trading manually, you will override your algo and revenge trade. Instead of FOMO-buying a stock, you will add a new "momentum" strategy to your algo suite mid-week because Nifty moved 3% and your current algo missed it. Instead of overtrading on expiry, you will "just run a quick manual trade" alongside your algo "to capture the move."

The behavioral patterns are in YOU, not in your trading interface. Changing the interface from a manual order window to a Python script does not change the human sitting behind it.

PortoAI connects to your Zerodha or Groww account and reads your actual trading data. It does not care whether you placed the trade manually or through an API. What it measures is the pattern: are you trading more frequently after losses? Are your position sizes increasing during drawdowns? Are you abandoning strategies after short losing streaks?

These patterns show up identically in manual traders and algo traders. The algo trader just has a more sophisticated story about why this time is different.

What should you actually do this week?

If you are an algo trader, here is a concrete checklist for the post-April 1 world.

Compliance first. Set up your static IP, configure daily auto-logout, enable 2FA on your broker API. Do not put this off. Non-compliant orders may be rejected by the exchange without warning.

Audit your override history. Go through the last 6 months of your trading data. How many times did you manually intervene in your algo's trades? What was the P&L of those interventions versus what would have happened if you had not touched it? In most cases, the interventions cost you money.

Calculate your true cost. Cloud server, static IP, API fees, broker charges, STT (which just increased 150% on futures). If your algo's gross return does not clear these costs by at least 2x, the strategy does not have enough edge to be worth running.

Set override rules. Before the market opens Monday, write down the specific conditions under which you are allowed to override your algo. "I feel uncomfortable" is not a condition. "Maximum drawdown exceeds 15% of allocated capital" is. If the condition is not met, you do not touch the system. Period.

Connect PortoAI. Even if you trade algorithmically, your behavioral fingerprint tells you whether your interventions are helping or hurting. The data does not lie. Your backtest might. Your gut definitely does.

Frequently Asked Questions

What are the new SEBI algo trading rules from April 2026?

From April 1, 2026, every algorithmic order on Indian exchanges must carry a unique Algo-ID registered with the exchange through your broker. All API access requires OAuth-based 2FA, static IP addresses, and daily auto-logout. Black box algo providers must hold a SEBI Research Analyst license and disclose performance data. Brokers carry full legal responsibility for every algo running through their systems. If your algo places more than 10 orders per second on any exchange, it must be formally registered.

Do SEBI algo rules apply to retail traders using Zerodha or Groww APIs?

If you use the Zerodha Kite Connect API or any broker API to place automated trades, yes. The rules apply to any order generated by code rather than manual input. However, if your trading frequency stays below 10 orders per second, you do not need to register your strategy with the exchange. You still need to comply with static IP, 2FA, and daily logout requirements. If you only trade manually through the app or web terminal, these rules do not affect you.

Can algorithmic trading make you a profitable trader in India?

No technology guarantees profitability. SEBI data shows that 93% of individual F&O traders lost money between FY22 and FY24, and algo traders are not exempt from this statistic. The advantage of algorithmic trading is consistency: it removes in-the-moment emotional decisions. The disadvantage is that the human behind the algo can still override it, abandon it during drawdowns, or over-optimize it to fit past data that will not repeat. Profitability depends on having a genuine edge in your strategy, not on whether a bot executes it.

Why do algo traders still lose money despite automation?

Three patterns account for most algo trader losses. First, over-optimization: fitting your strategy to historical data so precisely that it fails on live markets. Second, override behavior: manually intervening when the algo hits a drawdown, turning off the system at exactly the wrong moment. Third, strategy hopping: abandoning an algo after 2-3 losing weeks and switching to a new one, never allowing any strategy enough time to prove its edge. The bot is emotionless. The person controlling the on/off switch is not.

What is the 10 orders per second threshold for algo registration?

SEBI set a threshold of 10 orders per second per exchange within any calendar second. If your algorithm stays below this rate, you are not required to formally register your strategy with the exchange. You must still comply with all other requirements: static IP, 2FA, OAuth authentication, daily auto-logout, and routing all orders through your broker. Most retail algo traders operating on Zerodha Kite Connect or similar APIs trade well below this threshold.

How does PortoAI detect algo override behavior?

PortoAI connects to your Zerodha or Groww account via read-only API and monitors your trading patterns over time. When you run an algo, your trade frequency and timing are consistent: similar number of trades, similar entry and exit times, predictable position sizes. When you override the algo and start trading manually, the pattern breaks. PortoAI's behavioral fingerprint system detects this shift, flags it as a potential override, and sends a cooling period alert before your next manual trade.

Is algo trading better than manual trading for F&O in India?

Neither is inherently better. The method of execution is not the variable that determines profitability. What matters is whether you have a genuine edge (a strategy that produces positive expected value after all costs) and the discipline to execute it consistently. An algo removes in-the-moment emotion from individual trade execution. It does not remove emotion from strategy selection, parameter tuning, start/stop decisions, or the choice to override. If your behavioral patterns are the reason you lose money, an algo relocates the problem rather than solving it.