How it works
Financial news tells you an event happened. It rarely tells you the thing that matters: who else is affected, and how? The obvious effect is priced within minutes of the headline. The interesting one is two or three steps down the chain, hitting a company nobody has mentioned, or a currency, or a fund forced to sell something unrelated.
MarketLens traces those chains, every weekday morning, in the same three stages.
The three stages
| Stage | Question it answers |
|---|---|
| 1. What happened | The event in plain English, and whether the market had already expected it. |
| 2. How it spreads | Each consequence, link by link, with each link caused by the one before it. |
| 3. What it means | The effect on each market, with a direction, a size, a timeframe and a confidence score. |
Picking the stories
Every morning before the London open, the system reads headlines from 23 sources: central banks and statistical agencies, wire services and financial newsrooms, and specialist feeds for each asset class. Around 257 articles come in on a normal day.
Near-identical headlines are grouped, so a story carried by eight outlets counts as one event with eight sources rather than eight events. Each group is then scored out of 100:
| What is measured | Max | Why it counts |
|---|---|---|
| How many outlets ran it | 30 | The best available signal that something is genuinely important |
| How authoritative the source is | 20 | A central bank announcement outranks an aggregator |
| How many markets it touches | 20 | Breadth is what makes a story worth a full analysis |
| How market-relevant the language is | 20 | Weighted vocabulary, so "tariff" counts for more than "quarterly" |
| How fresh it is | 10 | Decays over a 36-hour window |
This step is deliberately mechanical and cheap. It cuts several hundred headlines down to about 25 before any expensive analysis begins. The score for each published story is shown on its own page, so you can check the working.
Tracing the chain
Every knock-on effect has to travel through one of eight named routes. That constraint is the whole design. Without it, this kind of analysis drifts into saying markets might be volatile, which is true every day and useful never.
| Route | What it means |
|---|---|
| What central banks do next | Changes how likely it is that central banks cut or raise interest rates, and how quickly. |
| The cost of money | Moves government bond yields, which set the baseline return every other investment is judged against. When that baseline moves, everything reprices. |
| Company profits | Changes revenue, costs or pricing power somewhere in a supply chain, including for companies not mentioned in the story. |
| Borrowing costs | Changes how expensive or how easy it is for companies to borrow, which matters most for those already carrying a lot of debt. |
| Currencies and trade | Moves an exchange rate, which changes what importers pay and what exporters earn. |
| Who is forced to trade | Interacts with bets investors already hold. When a crowded position goes wrong, forced selling pushes the move further than the news alone justifies. |
| How assets move together | Changes whether assets that normally offset each other still do. When those relationships break, hedges stop working and leveraged funds are forced to cut risk. |
| Knock-on to similar assets | Read-across to competitors, suppliers, customers and assets that investors treat as alternatives. |
Two versions of everything
Each claim is written twice. The plain version explains what it means to someone who reads the news but does not work in markets. Underneath sits the same point in the language a trading desk would use, with the exact mechanism. Neither reader has to put up with the other's version.
Being marked
Whenever the analysis makes a confident directional call, it is logged with the market price at that moment and scored against what actually happened once its timeframe is up. The scorecard shows the record, including every miss, broken down by market and by how confident the call was. If the confident calls are not more accurate than the hedged ones, the confidence scale is meaningless, and that will be visible on the page rather than hidden.
What it cannot do
- The analysis is generated by a language model. It reasons well about how things connect and badly about precise numbers, which is why every size is a range and no figure is presented as a data point.
- It reads only freely available sources. Paywalled reporting and real-time wire services are not included.
- Market data is delayed. It is used for context and for scoring, never for trading.
- It runs once a day. This is a way of thinking about how news travels, not a live trading signal.
Who built it
Jonathan Savill. MarketLens is an independent project, built to develop and demonstrate a structured way of reading cross-asset news. The whole pipeline, from reading the feeds to scoring the calls to rendering this page, is custom-built with no third-party code.