Editorial desk · Taipei City · free to read +886 2 2727 5418
Echo Investment Journal

Coverage

Coverage guide

The journal's editorial map: five threads of machine-read financial news, the checks behind each one, and how a topic turns into an article. Free to read, ordered by curiosity rather than by a calendar.

Signal fundamentals

Where tone comes from before any model runs.

  • Loughran–McDonald word lists, and why finance needs its own
  • VADER-style scoring engines and their blind spots
  • Annotated news corpora — who labels the text, and how they disagree

Machine reading

How modern models read sentences, not words.

  • FinBERT-family classifiers: training data, versions, drift
  • Named-entity recognition on company mentions
  • Multilingual news — tone across languages and wire services

From headline to index

The arithmetic between one article and one number.

  • Aggregation weights: story volume, recency, outlet reach
  • Headline-versus-body scoring, and what each predicts
  • Published sentiment indexes — how to read their moving parts

Honest testing

What makes a sentiment claim testable rather than decorative.

  • Backtest design: windows, costs, turnover
  • Overfitting stories, and how they get exposed
  • Academic findings translated into working English, including Tetlock's 2007 evidence on media tone and markets

Failure modes

Where readers get burned if they trust the read blindly.

  • Sarcasm, irony and the recycled-alert problem
  • Boilerplate language in filings and rating actions
  • Press releases and bot output scoring as if they were news
Taipei 101 and the Xinyi District skyline at night, seen against a dark sky.
Xinyi District after dark. The journal's desk sits in this part of Taipei City, reading the day's wires and checking what the machines said about them.

How a topic becomes an article

  1. A concrete question. Each piece starts from a question narrow enough to answer: what does this index actually measure? — never just a mood.
  2. Rebuild from primary sources. The model card, the paper, the filing, the newsroom's own description. Second-hand summaries are used for direction, not as evidence.
  3. Run the five checks. The gate from the front page — source, model, baseline, granularity, failure. Anything unprovable is labelled exploratory, not established.
  4. Publish free, invite challenge. Articles stay free to read; disagreements reach the desk and are answered where other readers can see the exchange.

Bring a question to the desk. Coverage that answers reader questions beats coverage that answers a content calendar — the inbox is where threads get sharpened.