Congratulations! That feeling of watching your own system
execute trades automatically for the first time is unlike
anything else.
50% win rate with positive P&L in week 1 is actually a
solid start — most new algos either overtrade or miss
entries in the first week while parameters settle.
One thing worth tracking during your 30-day paper period
beyond P&L — position reconciliation. Check regularly
that what your system thinks it holds matches what the
broker actually shows. In live trading, order rejections
and partial fills can silently diverge your system state
from broker state. Better to discover that behaviour in
paper mode.
Also, paper results tend to be slightly better than live
because slippage is not simulated realistically. When you
go live, expect the first 2 weeks to feel worse than paper
even if the strategy is identical.
What stack are you using — Kite Connect or something else?
Thanks!
Yeah in paper mode there’s actually no broker-side state yet since fills are processed locally inside the engine itself.
About reconciliation, last time while testing directly with broker by placing orders sl tp canceling rejection manual close etc it worked fine, but i will monitor when i deploy this strategy with real captial after 20-30d.
Also, I’ve currently turned slippage OFF during paper runs because fills are happening on real arrived ticks.
But during backtests I already modelled 0.1% slippage + Indian market charges to keep expectations more realistic before live deployment. And i place SL tp at broker side so delay is not much issue, will check in live deployment though
Current stack is Python + Nautilus Trader with AngelOne Smart API integration. Wanted proper backtest → paper → live parity instead of the usual webhook chain setup.
Nautilus Trader is serious infrastructure — most retail
algo builders never go that deep. Backtest → paper → live
parity is the right way to build this, even if it takes
longer upfront.
Real-tick paper with fills on arrived ticks is actually
cleaner than most setups. The 0.1% slippage model in
backtest is realistic for NIFTY options during normal
liquidity — though on high-VIX days spreads widen
significantly and 0.1% can be optimistic on OTM strikes.
Broker-side SL/TP is the right call. The alternative
of managing exits in application state adds a whole
layer of failure modes.
One thing to watch with AngelOne Smart API — their
WebSocket feed occasionally drops during high-volatility
periods. Worth building a reconnection handler with
state recovery if you haven’t already, so a dropped
connection doesn’t leave an unmonitored open position.
I built on Kite Connect — different API quirks but
similar challenges. The position reconciliation issue
I mentioned earlier caught me off guard in live —
order rejections that returned success codes initially.
What strategies are you running — directional or
non-directional?
AngelOne stability has improved — good to hear last
month has been clean. The WebSocket drops I saw were
during high-volatility events so hopefully less frequent
now.
2 lakh LOC is serious infrastructure. Are you building
a platform for multiple strategies or keeping it
single-strategy focused?
yeah its multi strategy, currently 2 are deployed for paper trading
here i can backtest all type of strategy and deploy all at once, each segment sebi rules charges etc everything
only if i had tick level data i was planning for hft (not like institutions) but more number of trades, so storing ticks also for future backtest
here is 2nd strategy it is high win rate low frequency, monthly 1-2 trades only
For stocks u can get data on kaggle, around 500 stock data set is there from 2015 i guess, and n50 from 2000 i guess, for fno i took data from dhan u get 5y of data 1min & 10stirke price atm and itm
For nautilus there are no learning resource cause no one in india doing python based algo trading seriously, codex and claude are the only option for learning and guidance
claude can help write code or guide, but u have to make decision, architecture, data flow, decision flow etc
i have kept everything seperate, decision logic, signal logic, execution logic, analytics logic and so on
i sugguest, if ur not developer go with quantconnect lean, they have huge community zerodha support, but it is subscription based
i choose nautilus bcoz of speed and streaming data that was my main concern without it my 16gb ram will get out of ram
if u need any help contact me insta i am more active there
Clean dashboard — the Rules Compliance panel with daily
loss limit and max drawdown limits built in is exactly
the right architecture. Most retail algos skip that layer
and pay for it eventually.
High win rate, low frequency (1-2 trades/month) is an
interesting approach — lower execution risk, less slippage
exposure, but drawdown periods between trades can test
patience. How are you handling position sizing on a
monthly-frequency strategy — fixed capital per trade or
volatility-adjusted?
100% win rate on 1 trade is a good start. The real test
will be at 20-30 trades when the distribution starts
to show.
Tick storage for future backtest is smart — most people
realize they need it only after they needed it.
ETF momentum on gold and silver makes sense for low-frequency
high win rate — those markets trend more cleanly than indices
and have less intraday noise to fight through.
% based capital sizing is the right call for ETFs given
the lower volatility profile compared to F&O.
Will check out your Instagram — what’s the handle?
not intrested to sell my framework nor strategy, and see our hardwork selling on telegram
yeah but will try how to maximize the revenue with good ways but always main source will be trading i guess
majority of people do no code algos like streak, so in india there are few people who do python algo trading like u, working around few years but due to lesser community we met today if we had met earlier u could have helped me lot, instead of learning from my own mistakes
I do the executions myself via python for non-directional OS strategies (with some directional elements). Backtesting is done using tradetron/algotest etc combined with some smart excel for added edge.
For directional (Nifty OB, Gold/Silver Fut) I only use pine script (TV) and webhooks.
Non-directional STBT strategies are not that susceptible to curve fitting, a lot of it is basically risk balancing. But directional positional strategies are very prone to curve fitting. Important to have large sample size, run sensitivity tests and do some forward testing.
tradetron/algo test take subscription for each backtest right? and they are no code platform right?
webhooks add delay if timing matters, pine script have less logic not versatile as pythong so i decided to build whole framework where i can backtest paper and deploy all type of strategy
added params optimization to find best param but i run this on 6month only to avoid overfitting, use wfo on remaining months
and how do u handle order slip? for fno huge change have happed from last 2y due to expiry change and lot size so i dont much consider previous data than that, only for signal testing
i have observed one more thing, in signal testing if get 60-70%+ win rate, actual win rate after placing options trades it came down to 50% only due to decay,
so what i was thinking to switch to naked short selling, looks like u have experince in that can u share ur insigts with me? any case where sl tp got slipped how to handle and risk came with shortselling?
i would like to connect directly, isnta twitter? or any other platform?
Yes. I use them only for backtesting, executions I do by myself (python) - because I want control, also because I have 1-2 additional filters on top.
Yes, few seconds sometimes, and my strategies are candle close 15/60 min, so 2-3 seconds delay is acceptable overall.
6 months cycle param optimization for directional positional strategies? It’s something I would never do. Fine-tuning too much is bad in my book.
I backtest on the options premium only for non directional strategies, and 2022-present is kind of enough for those type of strategies. You can backtest based on DTE, and lot size changes are not a problem.
I think win-rate differences are not that high, but ya, returns are reduced. I dont set small targets and sl, especially since my OS strategies have dynamic multi legs that keep rolling, I’d be busted by slippages if I set small tgts sl.
Btw if you take decisions in the 1st min, large divergence from backtest is inevitable.
Signals triggering correctly
