What is trading?
CROSS-CUTTING: COURSE 1 | LESSON 4
Macro and news digestion workflows
Learning objectives
Build a repeatable "paste-then-digest" workflow for economic calendars, central bank statements and news.
Use diff-style prompts to extract what actually changed between two policy statements.
Apply the source-verification pass that keeps digested research from smuggling hallucinations into your plan.
The problem: too much text, too little time
A serious trading week generates a reading pile nobody finishes: the economic calendar, two or three central bank communications, rate decisions, inflation prints, a dozen analyst notes. Most traders respond in one of two bad ways — skip it all and trade blind, or skim headlines and mistake vibes for analysis.
AI assistants are genuinely excellent at this problem, with one giant caveat that should be reflexive by now: the model has no idea what happened this week until you paste it in. A knowledge cutoff (Lesson X1.1) means "summarise this week's macro picture" without pasted material produces a confident essay about some past week, dressed as the present. The whole workflow rests on one rule:
You are the data pipeline. The AI is the processing plant.
Workflow 1 — the morning calendar briefing
Each morning (or Sunday for the week), copy the day's rows from a reliable economic calendar — time, currency, event, forecast, previous — and paste them with a digestion prompt:
That last section is a deliberately clever prompt trick: instead of letting the model invent events (hallucination), you channel its background knowledge into a checklist for auditing your own export. Its memory becomes a safety net rather than a fact source.
What this briefing is not: a forecast. "CPI is the day's main volatility risk" is scheduling information. "CPI will come in hot, buy dollars" is a prediction — and if your AI produces sentences like that, your prompt is missing its guard rails.
Workflow 2 — the statement diff
Central bank statements are written to be compared. Markets move on changes between one statement and the next — a dropped phrase, a softened adjective. Diffing two statements by eye is miserable; for an LLM it's a core competency.
Paste both statements (current and previous, from the central bank's own website — not a news summary of them) and ask:
Then run the grounding check from Lesson X1.2 on two or three of the quoted changes: search the pasted documents for the exact wording. If a "quote" isn't there verbatim, discard that row and downgrade your trust in the rest. When quotes check out, you've compressed forty minutes of squinting into five minutes of verification.
The same diff pattern works on inflation report summaries, rate-decision press releases, and your own broker's margin-requirement notices before big events.
Workflow 3 — the article distiller
For analyst notes and news articles, digest with role separation between reporting and opinion:
Section 2 matters most. News digestion goes wrong when a journalist's forecast ("analysts expect the pair to test 1.10") launders itself into your head as a fact. Forcing the fact/opinion split keeps forecasts labelled as somebody's opinion — which is all they ever were. And section 3 trains you to notice what a persuasive article conveniently skipped.
The verification pass, adapted for digestion
Digested research feels safer than raw AI generation because "the AI only summarised". Don't relax. Summarisation has its own failure modes: dropped negations ("not expected to cut" → "expected to cut"), merged numbers, and invented smoothing between bullet points. Before anything reaches your trading plan:
- Numbers: every figure in the digest exists in the pasted source, unchanged.
- Quotes: verbatim-searchable in the source.
- Negations: for any summary line that would change your bias, reread the original sentence. Two minutes, and it's where the worst errors hide.
- Recency: the source you pasted is actually current — the AI can't check the date's plausibility for you.
Ten minutes total for a morning briefing plus a statement diff, versus an hour of unstructured reading — that's the honest value proposition. Not better predictions. More organised inputs, faster, with the thinking still yours.
The reusable versions
The skills library packages these workflows: Morning Macro Briefing (skills/morning-macro-briefing.md) for workflow 1 and Economic Event Explainer (skills/economic-event-explainer.md) for understanding individual releases in depth. Copy once, paste data daily.
Key takeaways
The model doesn't know what happened this week — you paste the calendar, statements and articles in; it organises, never sources.
Statement diffs ("list every change, quote exactly") are the highest-value macro prompt: markets move on changes, and LLMs diff text superbly.
Channel the model's stale memory into audit checklists ("what's commonly relevant that's NOT in this data?") instead of letting it state facts.
Force the fact/opinion split when digesting articles so forecasts stay labelled as opinions.
Verify numbers, quotes and especially negations against the pasted source — summarisation errors are quieter than hallucinations but just as costly.