News Impact Scorer

Economic calendars (Investing.com, Forex Factory, Trading Economics) for economic data. Company investor relations pages and SEDAR+ for earnings dates. Bloomberg/Reuters calendars for comprehensive coverage. Set alerts for your watchlist events.

For most traders, no. Initial reactions are volatile and can reverse. Consider waiting 15-30 minutes for dust to settle. Professionals with fast systems can trade immediately, but retail traders often benefit from letting initial reaction play out.

Analyst consensus estimates are available on financial sites (Yahoo Finance, Reuters, Bloomberg). Look for EPS and revenue estimates. Also check the 'whisper number' - sometimes expectations are higher than published estimates.

Several reasons: news was less bad than feared (positive surprise), bad news was already priced in (sell the rumor, buy the news), short covering on news (shorts exit), or market focuses on one positive element in otherwise bad news.

Significantly. US is Canada's largest trading partner. Fed decisions affect global rates and CAD. US economic data affects Canadian exporters and risk sentiment. TSX often follows S&P 500's reaction to US news.

Collect: actual EPS from filings/reports, estimate consensus from data providers (IBES, Bloomberg), calculate surprise. Store with date, stock, actual, estimate, surprise %. Match with subsequent returns. Build over time for pattern analysis.

Scheduled: known timing allows positioning, but market already braced, may be priced in. Unscheduled: surprise timing creates gaps but also more genuine price discovery. Unscheduled often has larger initial move but scheduled may have clearer patterns.

Start with: Event type 40%, Surprise 30%, Context 20%, Timing 10%. Backtest and adjust. Some find surprise matters most; others weight context heavily. Your weights should reflect what predicts returns best in your backtests.

When a sector leader reports, analyze results for industry implications. If RY's Canadian banking is strong, position long TD or BMO before their reports. Risk: each company has idiosyncratic factors. Size smaller than direct trades.

Combine: technical gives support/resistance for entry/exit; news gives directional catalyst. Strong news + breakout above resistance = higher conviction. News counter to technical trend = be cautious. Use news to time within technical framework.

Pipeline: 1) Ingest news from APIs/feeds, 2) Preprocess (clean, tokenize), 3) Apply model (FinBERT or custom), 4) Extract entities/events, 5) Score sentiment and magnitude, 6) Link to securities, 7) Store and feed to trading system. Use Python with transformers, spacy.

Competitive HFT: sub-10 milliseconds from news to order. Systematic but not HFT: 100ms-1 second is acceptable. Discretionary with automation: seconds is fine. Match latency to strategy - fast for fading initial move, slower for trend following post-news.

Critical: use timestamps of when news was actually available (not price timestamps). Include latency assumptions (when could you realistically trade). Use realistic slippage (news events have wide spreads). Walk-forward testing essential as news dynamics change.

Map relationships: stock ↔ sector, stock ↔ currency, macro ↔ rates ↔ equities. Estimate lead/lag relationships from historical data. When news hits, score primary impact, then propagate through network with appropriate weights and delays.

Use walk-forward validation, not in-sample optimization. Keep model simple - few parameters. Test on different time periods and markets. Use regularization in ML models. Sanity check - does the model make intuitive sense? Avoid data mining without hypothesis.