forecasting-ch20-marketing-mixlisted

# Chapter 20: The Marketing Mix ## Scope and intake Use when media carryover, response saturation or attribution stability must be assessed against business sales. Do not use for confusing prelaunch trial calibration with retrospective MMM, or certifying causal ROI from predictive fit. Ask only for unresolved material inputs: Is the question predictive sales, incremental effect or launch volume? What pre-window spend exists? Which demand drivers affect both spend and sales? What experimental calibration or defensible external priors exist? ## Input contract and additional evidence A regular series of `sales` with one nonnegative spend column per channel and any numeric control columns (weather, price, competitor activity). Name the channels in config `channels` and the controls in `controls`. The older two-channel example (`spend_a,spend_b` with fixed `decay_a`, `half_a`, ...) still runs unchanged. Minimal **format illustration**, not sufficient training data: ```csv timestamp,sales,tv,digital,radio,temp 2010-01-01,131.2,54.0,22.5,8.1,-0.4 2010-02-01,128.7,61.3,19.0,12.6,0.2 2010-03-01,140.9,12.8,70.4,3.3,1.1 ``` ## Executable interface Exact CLI columns: `timestamp,sales,<channels...>[,controls...]`. Supported method controls: `alpha, as_of, channels, controls, decay_a, decay_b, decay_grid, frequency, half_a, half_b, horizon, initial_a, initial_b, noise_sd, origins, prior_mean, prior_sd, reallocation_total, saturation, season, seed, windows`. For every channel the