Signals board

Africa still sells travel through agents. The agent's hardest daily question is not what the fare is, it is whether to issue the ticket now or wait, and on a Lagos to London return that decision is worth more than the commission on the ticket. FareRadar scores that decision for 25 Africa-origin routes across the next 180 departure dates and resolves each one to issue, hold or watch.

Where the prices come from

The prices on this site are generated by a structural model, not pulled from a live GDS. They are realistic in shape and magnitude because they are built from the published economics of each route, but they are not quotes and no seat can be bought at them. Treat the site as a demonstration of the decision layer, not as a fare search.

The model composes five terms:

• Filed base fare per route, set from typical published economy return levels in each origin market.
• Market seasonality, a monthly multiplier per route class. Diaspora long-haul, Gulf, intra-Africa and pilgrimage routes peak in different months and the multipliers reflect that.
• Demand events, a calendar of the windows that actually move African fares: the September student intake, GITEX, US Thanksgiving, Detty December, the Dubai Shopping Festival, the Umrah build-up, and the January collapse. Each ramps in and out over three days.
• The booking curve, cheapest around 60 to 90 days out and climbing steeply inside three weeks.
• Day of week, with Friday to Sunday departures carrying a premium.

The observed fare is that expectation plus a mean-reverting noise process, with occasional sustained shocks standing in for a fare sale, a competitor seat dump, or cheap buckets closing out. The process is seeded from the route and departure date, so the same date always produces the same price and the board does not reshuffle between page loads.

How the signal is scored

Two numbers come out of every route and date:

• Distance from band, how many standard deviations today's price sits from its typical level for that date and that point on the booking curve.
• The 14-day view, where the booking curve puts the fare two weeks from now, assuming about a third of today's dislocation persists rather than correcting immediately.

Issue when the fare is more than 0.8 standard deviations under band, or the 14-day view is up 10% or more. Hold when it is trading above band with more than 45 days of runway and a flat curve. Watch otherwise. Confidence scales with the size of the dislocation. Value at stake is the per-seat gap between today's price and the 14-day projection, which is the number the decision actually turns on.

What it would take to run this on live data

The analytics layer never talks to a fare source directly. It asks an adapter for a price for a route and a departure date. A live adapter for Amadeus Flight Offers Search ships with the code and activates on credentials. The one real gap is history: search APIs return today's price and nothing else, so a production deployment polls on a schedule and accumulates its own price series. The scoring engine does not change either way, because it only ever needs a series of observation date, days to departure, and price.

Honest limits

• The typical band is derived from the same process that generates the price, so the site demonstrates the decision logic rather than proving the model against the market.
• Fares are one adult, economy, cheapest available. No cabin mix, no fare rules, no baggage, no change penalties, all of which matter to a real agent's margin.
• Seat availability is not modelled at all. A cheap fare with two seats left is a different decision from a cheap fare with forty.
• FX is a fixed table, not a live rate, which matters more than usual in the naira.
• The event calendar is hand-built and would need maintaining per season in production.