ATTRIBUTION MODELS, MEDIA BUDGET ALLOCATION, AND LOSS IN CFD BROKERAGE

Abstract

Introduction. Paid customer acquisition is the dominant cost line for retail contracts-for-difference (CFD) and forex brokers, and the efficiency of that spend is judged through an attribution model that assigns conversions to marketing channels. The standard convention in retail brokerage is last-click attribution anchored to the registration (signup) event, after which the registration source is treated as the owner of all subsequent revenue.

Purpose. The study quantifies the economic cost of this convention. It builds an economic-mathematical model of how last-click attribution combined with signup-source lock-in biases the estimated cost of acquisition (CAC) across channels and misallocates the paid media budget, and it expresses the resulting loss in monetary terms.

Materials and methods. The model formalises a forex acquisition event ladder (signup, identity verification, first-time deposit, first trade, recurring deposits and trades) and two distinct layers of distortion: touch distortion at the signup event and source lock-in across all post-signup revenue. The true contribution of each channel is approximated by the Shapley value computed over a reproducible converting-journey coalition structure. Channel response is modelled with diminishing returns and inventory ceilings; the loss is the contribution margin forgone when the budget is optimised on locked signals instead of true contribution. Parameterisation is calibrated on an aggregated, anonymised portfolio of seven EU Tier-1 CFD and forex broker acquisition projects run by the author over two years.

Results. Under the lock-in convention, closing channels (brand search and direct) capture credit far above their Shapley contribution, while initiating and post-signup activation channels are systematically under-credited or rendered invisible. Optimising the budget on these biased signals forgoes between roughly three and eleven percent of the paid budget in contribution margin, and the loss rises as the lag between signup and deposit lengthens, because a longer lag leaves more uncredited reactivation work. Discounting of delayed revenue adds a small further adjustment.

Prospects. Further work can extend the model to redeposit and trading-volume horizons, to incrementality-based truth proxies, and to multi-market parameterisations.