Meet PSX Copilot — your AI stock assistant
Ask anything about the Pakistan Stock Exchange in plain English. PSX Copilot answers instantly using live market data and our forecast model trained on every stock's full history — completely free.
…or tap the “Ask AI” button in the bottom-right corner.
What you can ask
“Should I buy OGDC?”
Get the model's signal, confidence, price target and the stock's own backtested accuracy.
“What's the RSI of LUCK?”
Live technical indicators — RSI, MACD, moving averages, ADX and a technical score.
“Fundamentals of HBL”
P/E, EPS, dividend yield, market cap and a fundamental score.
“Top buy ideas”
The strongest current buy and sell signals across the market.
“How is the market today?”
KSE-100 level, market breadth and overall forecast hit-rate.
“Compare ENGRO vs FFC”
Side-by-side scores, valuations and the model's preferred pick.
How it works
PSX Copilot is not a generic chatbot. It is a purpose-built assistant that understands your question, identifies the stock(s) involved, and responds using our own database of live PSX prices, technical indicators and company fundamentals, together with our logistic-regression forecast model trained on the full history of every stock. Answers are information and education only — never financial advice.
What it cannot do
Every tool of this kind has hard limits, and knowing them is the difference between using it well and being misled by it. Ours are:
- It does not know the news. The model reads prices, indicators and fundamentals. A new contract, a court ruling, a change of management or a flood at a factory reaches it only after the price has already moved.
- It cannot value a business. It matches statistical patterns in history. It has no opinion on whether a company is well run or whether its industry has a future.
- It is weakest where data is thin. For small, illiquid PSX listings, the price series reflects a handful of trades. Indicators computed on that are noise with a decimal point, and the model inherits that.
- It assumes the future rhymes with the past. It was trained on history. In a genuine regime change — a policy shock, a currency crisis — that assumption breaks exactly when you most want an answer.
We publish the model's real out-of-sample accuracy — including the periods it read badly — on the model performance page. It is not a spectacular number, because honest numbers on hard problems are not spectacular. If you want the fuller argument, read what technical indicators cannot tell you.