marketdx 0.5.1

The MarketDX financial impact graph, in Python — news→market impact (why, who, direction, ripple) across stocks, commodities, FX, crypto, and private companies.

marketdx

The financial impact graph, in Python.
News → who it touches and why — the causal channel, the story's lean, and the ripple — across stocks, commodities, FX, crypto, and private companies.
A research, screening & feature layer — direction is the news's content lean, not a price forecast.

▶ Live playground · Docs & pricing · Sample dataset

---

pip install marketdx          # add [pandas] for .to_df():  pip install "marketdx[pandas]"

An API key is required. Create a free one at https://marketdx.lab.ai (sign in → API keys), then pass it to the client (keep it out of source control — read it from an env var / secret in real apps):

from marketdx import MarketDX

mdx = MarketDX(api_key="avn_live_…")                     # your key from https://marketdx.lab.ai
for s in mdx.news(megatrend="ai-power", impact="indirect"):
    print(s.title, [(e.name, e.impact.net_direction) for e in s.entities])

That's the whole graph: every news event, every affected entity, labeled with direction, relevance, the causal aspect (the why), and whether it's the epicenter or a ripple — across five asset classes, including private companies ticker feeds can't see.

No key yet? Explore everything with zero signup in the playground, then grab a free key at https://marketdx.lab.ai. Every request is authenticated with your key (Authorization: Bearer <key>) and metered in credits.

Why the SDK (not just requests)

• Typed graph — signal.entities[0].impact.aspects[0].direction with autocomplete, not raw dicts.
• Auto-pagination — for s in mdx.news(...) walks every page for you. Nothing to manage.
• Names, not ids — megatrend="ai-power" or "AI Power & Cooling" or 10040000 all work.
• .to_df() — the whole result as a pandas DataFrame, one row per (event × entity × aspect).
• Typed errors — AuthError, QuotaError, RateLimitError, BadRequestError, NotFoundError.

The graph, a few ways

# 1. Ripple: a themed event that also touches NON-thematic entities (our differentiator)
for s in mdx.news(megatrend="semiconductors", impact="indirect", max_items=50):
    ...

# 2. Beyond tickers: private companies in a trend (OpenAI, Anthropic, Ampere, ChangXin…)
for c in mdx.megatrends("semiconductors").off_coverage():
    print(c.name, c.type, "→", c.megatrend["node_name"])

# 2b. …or just the NEWS that moves private companies — one server-side filter on the feed
for s in mdx.news(entity_type="private", only_scored=True):   # also: crypto / commodity / forex / stock
    print(s.title, [e.name for e in s.entities if e.type == "private"])

# 3. Per-stock impact timeline + its news-derived rivals
tl    = mdx.stock("NVDA.US").news(aspect="competition")
peers = mdx.stock("NVDA.US").competitors()

# 4. News-driven screen — where the news leans positive on a theme (the model's read, for research)
positive_lean = mdx.stocks(megatrend="ai-power", direction="pos", country="US", order_by="news_count")

# 5. Semantic search — match news by MEANING, not keywords
hits = mdx.news_search("chip export controls to China")

# 6. Theme brief in ONE call — pulse timeseries + top stories + winners/losers + heatmap + assets
brief = mdx.theme("ai-power").summary(window="qtd")     # 7d/30d/90d/1y or mtd/qtd/ytd
print(brief["pulse"]["story_count"], brief["pulse"]["net_direction"])
print([w["ticker"] for w in brief["winners"]], "vs", [l["ticker"] for l in brief["losers"]])
print([a["name"] for a in brief["top_assets"]])          # commodity / forex / crypto the theme moves

Straight to pandas

.to_df() returns the same columns as the sample dataset (impact-signals.csv) — so anything you prototyped on the free CSV runs unchanged on the live graph:

df = mdx.news(megatrend="ai-power", impact="indirect").to_df()
# published_at · title · brief_text · entity_name · entity_ticker · entity_type · direction ·
# aspect · reason · relevance · impact · impact_score · node_name · entity_country · publisher · url

df.groupby(["entity_type", "direction"]).size()      # who the news lands on, +/− by asset class (the model's read)
df[df.aspect == "tariff"].entity_name.value_counts()  # who the tariff channel touches

Filtering — read this before you filter

Feed filters are EVENT-level, not row-level. news(direction=…, aspect=…, news_type=…, country=…) selects articles that contain at least one matching impact and returns the whole article with all its entities and aspects. So direction="pos" can return an article that also moves something neg, and aspect="supply" can return one whose other entities are hit via monetary. For exact per-row filtering, post-filter the entities:

for s in mdx.news(news_type="commodity_supply", direction="pos"):
    for e in s.entities:
        for a in (e.impact.aspects if e.impact else []):
            if e.type == "commodity" and a.direction == "pos" and a.aspect == "supply":
                ...   # exact row you asked for

Two different "direct" axes (don't conflate):

• Signal.impact_type (direct|indirect) = the article's relation to the queried node — epicenter (direct) vs ripple (indirect). Set by news(megatrend=…, impact="indirect").
• Entity.direct (True|False) = whether that entity is factually mentioned in the article (True) vs impact-only / not named (False).

Narrow the feed by entity — server-side. news() (and news_search()) filter by entity_type / only_scored / min_relevance in the API, so page.total stays the exact filtered count (no wasted paging). These keep the whole article — for an exact per-row cut, still post-filter the entities as above.

mdx.news(entity_type="commodity")                      # feed → only stories that move a commodity
mdx.news(entity_type="private")                        # only stories moving a private co (OpenAI, SpaceX)
mdx.news(only_scored=True)                              # drop mention-only articles (keep judged impact)
mdx.news(min_relevance=0.8)                             # only a strongly-relevant scored entity
mdx.news(megatrend="ai-power", entity_type="crypto")   # entity filters compose with megatrend scope
mdx.news_search("oil supply shock", entity_type="commodity")  # search supports entity_type too
mdx.news_by_tickers("NVDA.US")                          # a covered STOCK's news (direct + indirect)
mdx.megatrends("ai-power").off_coverage()              # private / off-coverage roster

Entity filters do not apply to impact="indirect" (the ripple feed) — the API returns 400 if you combine them. On news_by_tickers, the ticker set already scopes the entities.

Only entities with a scored impact (many are mentioned-only) — only_scored=True narrows to such articles server-side; then read the scored entities off each signal:

scored = [e for s in mdx.news(megatrend="ai-power", only_scored=True)
          for e in s.entities if e.impact and e.impact.aspects]

stock(t).news() is a stock-centric timeline — a StockNews (the stock's own impact /trend/ relevance), not an entity graph (no entities[]). For the full graph of an article, use news().

Story-collapse (on by default). The same story is often republished / rewritten across outlets. news(), news_search() and news_by_tickers() merge those near-duplicates into a single signal by default (cosine-similarity grouping, server-side) so a feed reads one-story-one-row. Pass collapse=False when you want the raw, un-deduped stream — e.g. to measure coverage volume:

merged = mdx.news(megatrend="ai-power").to_list()                 # deduped (default)
raw    = mdx.news(megatrend="ai-power", collapse=False).to_list() # every republication

Theme brief — the whole picture in one call

mdx.theme(id).summary(...) (a theme = a megatrend node; also mdx.megatrends(id).summary(...)) returns a pre-composed analyst brief so you don't stitch 5+ requests together. It's a fixed composite dict, not a paginated list:

brief = mdx.theme("ai-power").summary(window="30d")     # 7d/30d/90d/180d/1y or mtd/qtd/ytd (or from_/to)
brief["pulse"]          # story_count, net_direction, pos/neg share + a `series` (volume+sentiment/bucket)
brief["top_stories"]    # epicenter, deduped; market-wraps & no-member-named stories deprioritized
brief["ripple"]         # indirect (ripple-in) stories, each with `via`
brief["winners"], brief["losers"]   # member stocks by net direction
brief["aspect_heatmap"] # which channels the theme is playing out through
brief["top_entities"]   # operating companies most in the news
brief["top_assets"]     # commodity / forex / crypto the theme moves (split out from companies)

Every count is story-deduped (20 outlets on one story = 1). The pulse.series is the momentum signal — there's no single momentum scalar (the latest bucket is the current, partial period). Cost: 15 credits.

Metering & errors

Every call carries X-Credits-Charged / X-RateLimit-*; check your balance any time (free, unmetered):

mdx.account()   # {'plan': …, 'credits': {'balance', 'daily_quota', 'resets_at', 'unlimited'}, 'rate_limit': …}

from marketdx.errors import QuotaError, RateLimitError

try:
    signals = mdx.news(megatrend="ai-power").to_list()
except RateLimitError as e:
    time.sleep(e.retry_after or 1)
except QuotaError:
    ...   # daily quota spent — resets 00:00 UTC

Enum values (aspect, direction, entity_type, …) are type hints for your editor — the API is the source of truth, so new values work without upgrading the SDK. The live list: mdx.enums().

Reference

news · news_search · news_types · megatrends (.stocks / .off_coverage) · gics (.stocks) · stocks (search + screener) · stock (.news / .competitors / .peers) · enums · account. Full API docs: marketdx.lab.ai.

License

MIT. Built by MarketDX — democratizing financial data.