Parakete
Rescans all 599 Borsa İstanbul stocks every night, combining financials, foreign ownership, disclosures, technicals and the macro regime into a single score, and expresses downside risk as a probability.
About the Product
Parakete is an analysis and monitoring terminal built for Borsa İstanbul. Every night it rescans all 599 stocks — price and technical indicators, balance sheets and income statements, foreign investor ratios, public disclosures, bank ratios and macro series — and combines them into a 0–100 score built from seven components weighted by business model. Its design principle is to give reasoning rather than signals: under every score it states in plain language which financial line, which disclosure and which technical level contributed how many points, and weekly score changes are attributed to components exactly. The machine learning model is trained on 174,000 feature vectors derived from 599 stocks across 8 years and answers a single question: will this stock fall into the worst-performing quintile over the next 90 days? The answer is a calibrated probability. By contrast the direction model does not separate from a coin flip (0.511 out-of-sample AUC), so the product never claims a stock will rise — and says so on screen. The BIST Copilot and AI Research layer never produces numbers; it interprets engine-generated tables with citations, limited to a whitelist of metrics. Radar, screener, stock detail, portfolio, sector, bulletin, alerts and risk management modules all share the same calculation path, and the iOS and Android apps deliver the morning bulletin and alerts as push notifications.
This is not investment advice. Parakete is an analysis and monitoring tool: it does not route orders, does not provide brokerage services and does not constitute investment advisory. The scores, probabilities and model outputs it produces are statistical estimates based on historical data and do not guarantee future returns. Responsibility for your investment decisions remains with you.
Key Features
Price and technical indicators come from Yahoo Finance (597 stocks, 15 years), financial statements from İş Yatırım (580 stocks, 20 quarters), daily foreign ownership ratios (1.19 million records), bank capital adequacy and non-performing loans from the Banks Association, disclosures from the public platform (6,082 records), and macro series from the central bank and Yahoo. Missing data is left blank rather than estimated.
Data layerThe Parakete Score is a weighted sum of seven components: broker settlement, fund flow, financials, valuation, technicals, insider transactions and news. Weights are not fixed — for a bank the balance sheet is the business itself so the financial component dominates (26) while valuation lightens (9) since EV/EBITDA is undefined; for a REIT the reverse applies.
ScoringConcrete lines such as net profit, net debt and equity are written out; for a bank the same place shows capital adequacy and loan-to-deposit ratios. Weekly differences are attributed exactly across components, and where data is absent the screen states it explicitly rather than hiding it.
ExplainabilityThe model is trained on 174,000 feature vectors from 599 stocks across 8 years and expresses 90-day downside risk as a calibrated probability. Out-of-sample AUC is 0.740 at 90 days, 0.729 at 180 and 0.687 at 365. The direction model sits at 0.511 — indistinguishable from a coin flip — which is why the product never predicts a rise.
Machine learningFinancials enter the calculation only after their actual disclosure date. Because targets overlap, a gap the length of the horizon is placed between training and test blocks; breaking this rule raises no error and merely inflates results, so the embargo is locked down by a dedicated test suite.
MethodologyA newly trained model first runs in shadow mode and reaches production only through manual promotion; the system raises a retraining warning when the model ages or the data distribution shifts. Predictions are frozen at publication and settled against real prices when the horizon closes — a live exam rather than retrospective optimism.
GovernanceThe BIST Copilot sits in the corner of every screen and explains a stock by combining its score, financials, risk probability and latest disclosures. AI Research breaks multi-step questions into tool steps, runs them on the deterministic engine, and lets the AI interpret only the resulting tables with citations. The metrics available to the AI are whitelisted, the numbers themselves always come from the engine, and every claim is linked back to its source step.
Artificial intelligenceRadar handles ranking, the screener filtering, stock detail the decision, the portfolio tracking and alerts monitoring. Risk management computes ATR-based stop distance and position size from capital and risk percentage, splitting the budget into a staged buying plan. The calendar gathers buybacks, rights issues and dividend dates in one stream, and every table exports as spreadsheet-compatible CSV exactly as displayed.
Daily workflowAdditional Modules
How It Works
- 01The nightly pipeline collects data
Prices, financials, foreign ownership, disclosures, bank ratios and macro series are pulled from their sources. Collectors are fault tolerant: if a source fails to respond that step is skipped quietly and the rest of the pipeline continues.
- 02The score is computed
Seven components are weighted by business model and reduced to a single 0–100 score, and each stock is also positioned against its peers by z-score within 25 canonical sectors.
- 03The model turns risk into probability
The machine learning service produces 90-day downside risk as a calibrated probability and lists the variables that drove it most, with readable labels.
- 04The reasoning is written
Concrete financial lines, relevant disclosures and technical levels are written under the score in plain language, and the weekly change is attributed component by component.
- 05It reaches screens and notifications
Radar, screener, portfolio and sector screens all share the same calculation path, while the morning bulletin and triggered alerts arrive as push notifications on web and mobile.
Who It's For
The free tier covers the daily radar, stock detail, score reasoning, disclosures, the dividend calendar and watchlist alerts.
The Pro tier unlocks the screener and saved templates, the portfolio and risk panel, machine learning downside risk, sector-neutral ranking, the morning bulletin and personal alert rules.
The enterprise tier adds cited multi-step AI Research, the model's forward-looking live scorecard and broker-level skill measurement.
Roadmap
Programmatic access to the score, risk and data layers for enterprise users — on the enterprise roadmap.
Team accounts and in-house user management, with pricing subject to discussion.
Technical Summary
- Coverage
- 599 BIST stocks; prices for 597 over 15 years, financials for 580 over 20 quarters
- Data sources
- Yahoo Finance, İş Yatırım, the Banks Association, the public disclosure platform, the central bank, TEFAS
- Licensed data
- Broker-level settlement data requires a paid exchange licence; it is never fabricated and is flagged as manual/demo on screen
- Score
- 0–100 from 7 components, weighted by industrial / bank / REIT schema
- Model performance
- Downside risk out-of-sample AUC: 0.740 at 90 days, 0.729 at 180, 0.687 at 365. Direction model 0.511 (coin flip) — the product makes no direction calls
- Mobile app
- iOS and Android; radar, bulletin, screener, portfolio, alerts and the Copilot on your phone. The morning bulletin and triggered alerts arrive as push notifications that open the relevant screen
- Security and compliance
- Email-verified registration and role-based access; the privacy notice, terms of use and disclaimer are published inside the product
Frequently Asked Questions
What is Parakete?
Rescans all 599 Borsa İstanbul stocks every night, combining financials, foreign ownership, disclosures, technicals and the macro regime into a single score, and expresses downside risk as a probability. This is not investment advice. Parakete is an analysis and monitoring tool: it does not route orders, does not provide brokerage services and does not constitute investment advisory. The scores, probabilities and model outputs it produces are statistical estimates based on historical data and do not guarantee future returns. Responsibility for your investment decisions remains with you.
Who is Parakete for?
Investors following the market: The free tier covers the daily radar, stock detail, score reasoning, disclosures, the dividend calendar and watchlist alerts. Self-directed individual investors: The Pro tier unlocks the screener and saved templates, the portfolio and risk panel, machine learning downside risk, sector-neutral ranking, the morning bulletin and personal alert rules. Institutions and research teams: The enterprise tier adds cited multi-step AI Research, the model's forward-looking live scorecard and broker-level skill measurement.
How does Parakete work?
1. The nightly pipeline collects data — Prices, financials, foreign ownership, disclosures, bank ratios and macro series are pulled from their sources. Collectors are fault tolerant: if a source fails to respond that step is skipped quietly and the rest of the pipeline continues. 2. The score is computed — Seven components are weighted by business model and reduced to a single 0–100 score, and each stock is also positioned against its peers by z-score within 25 canonical sectors. 3. The model turns risk into probability — The machine learning service produces 90-day downside risk as a calibrated probability and lists the variables that drove it most, with readable labels. 4. The reasoning is written — Concrete financial lines, relevant disclosures and technical levels are written under the score in plain language, and the weekly change is attributed component by component. 5. It reaches screens and notifications — Radar, screener, portfolio and sector screens all share the same calculation path, while the morning bulletin and triggered alerts arrive as push notifications on web and mobile.
Benefits
- Every number has a source; missing data is left blank, never estimated
- No look-ahead: financials enter only after their actual disclosure date
- Models that do not work are stated openly, not sold as signals
- Score changes are the exact sum of component differences, not a guess
Other Products
GeoEner — Investment & Maintenance Management System
A platform that digitalizes asset, investment and maintenance management end-to-end for electricity distribution and energy companies, with geographic information system (GIS) support.
Explore Product
GEONIVA — Drone and AI Grid Inspection
Generates drone missions from work orders, scans drone and field photos with AI, links each finding to the right grid equipment and writes approved findings into the maintenance form. Known in GeoEner as Inspection Management, it is the inspection module of the GEOENER family.
Explore ProductGeoEner PQM — Power Quality Management Software
Continuously monitors voltage sags, harmonics, flicker and unbalance; automatically audits against EN 50160 and regulatory power quality criteria, and produces audit-ready reports.
Explore ProductLet's start your digital transformation together
Reach out today to discuss your project with our expert team.