Product of Trend&Brand · Educational research
Analysis Methods, Engines & Data Sources
How PSX AI Analyzer collects public data and explains it using twelve specialist engines, source dates and transparent limitations.
Reading the Market Atlas
The eight map tiles group published company sector classifications for easier exploration. Banks includes banking-related classifications; Textiles combines textile-related classifications; Others contains the remaining companies. These groups are not official exchange sector indices. Tile areas are equal and do not represent market value.
Daily change is weighted by recorded positive company market capitalization, using only observations with a numeric daily percentage change. Coverage counts show how many companies qualified; missing observations are excluded rather than treated as zero. The alternate market-cap view adds available recorded capitalization. Selecting a tile updates its snapshot and company list. Calendar dates come from published corporate-event records; a book closure is not an inferred ex-date or an instruction to trade.
Collector agents and research engines
Five collector modules assemble market/history observations, company disclosures, corporate calendars, company announcement headings and macro context. They use public PSX pages and files, an exchange-rate API, and manually maintained macro series. Collection modules are software workers with defined responsibilities; they do not bypass protected sites or invent access to data.
Twelve independent engines examine price structure, financial performance, valuation sensitivity, risk, liquidity, distributions, corporate events, disclosures, sector peers, macro context, Shariah index membership and source quality. Existing chart, screen, comparison, correlation, backtest and portfolio tools retain their own calculations.
Specialist analyzer and chatbot
The analyzer uses deterministic calculations and stock-specific instructions for source-grounded AI narration. It is not represented as a separately fine-tuned model. A true fine-tune would require a licensed training corpus, validation examples and evidence that the resulting model meets quality requirements.
The chatbot explains the relevant engine observations in the user’s language. It must distinguish observed data, derived calculations and assumptions, mention missing or stale inputs, and avoid buy/sell/hold calls. AI wording can contain errors: source links and calculation methods remain available for verification.
Calculation limits
Technical indicators use recorded EOD bars. Wilder RSI describes recent price momentum; SMA and MACD summarize price history. Volatility uses the sample standard deviation of up to 60 log-return observations annualized by the square root of 252. Maximum drawdown is the historical peak-to-trough price decline over up to 252 sessions. These figures do not cap future losses.
Earnings scenarios multiply positive EPS by a peer-based multiple bounded to 4–20x for the central case. Multiple sensitivity changes that assumption ±20%; it does not forecast a price. P/E-derived TTM EPS is marked as derived. Source statement amounts are converted from PKR thousands, while EPS stays in PKR per share.
Coverage and corrections
Daily Quotations provides distributions, book closures and board meetings. Conflicting PDF rows are marked. Announcement headings and linked documents are preserved, but document contents are not claimed to have been read unless extraction is available. Cash-flow statements, debt exposures, auditor opinions and independent licensed news feeds may be missing.
Source publication dates and collection timestamps are shown separately. SBP figures may be manually maintained. Shariah index membership is a published classification, not a new religious certification. Use Contact & Feedback to report a wrong value with its source, symbol and date.