Tools › Finance and Business › Marketing › Cohort Retention and CLV › Example
Fader and Hardie 2007 Table 1: shifted-beta-geometric fit to the High End segment
| Tool | Cohort Retention and CLV |
| Source | Fader, P. S., and Hardie, B. G. S. (2007). "How to Project Customer Retention." Journal of Interactive Marketing, Vol. 21, No. 1, pp. 76–90. Table 1 and Appendix B. |
| Level | Graduate |
Problem
Of the High End customers acquired in a year, 86.9 % are still active after one year, then 74.3, 65.3, 59.3, 55.1, 51.7 and 49.1 % through year 7 (and 46.8, 44.5, 42.7, 40.9, 39.4 % through year 12). Fit the shifted-beta-geometric model to the first seven years by maximum likelihood, read the mixing distribution it implies, and project the retention to year 12.
Inputs to Enter
- Data layout None (fit the retention curve only); curve 1, 86.9 / 2, 74.3 / 3, 65.3 / 4, 59.3 / 5, 55.1 / 6, 51.7 / 7, 49.1 / 8, 46.8 / 9, 44.5 / 10, 42.7 / 11, 40.9 / 12, 39.4
- Fit through period 7; discount rate 10 %, horizon 12
Answers
- α = 0.668, β = 3.806 (Appendix B; the body text misprints 0.688), maximum log-likelihood −1.611; LL(1, 1) = −2.116 and LL(0.01, 0.01) = −2.742 as in the worksheet
- E(θ) = 0.15, reverse-J mixing distribution; projected S(12) = 37.80 % against the actual 39.4 % (about 4 %, an under-estimate, not the over-estimate the paper says)
- Geometric alternative θ = 0.1033 with LL −1.6371 (worse); the Berry–Linoff exponential regression quoted by the paper under-estimates year 12 by 30 %
Notes
Choose layout None and paste the High End column in the Curve box in place of the default Regular column; the holdout statistics use years 8–12 (MAPE about 3 %).