This is the table of contents for Week 9. There is no new Lab or Homework this week — the runway
is clear so you can finish Project 2. The interactives below are self-paced: you type a forecast, turn knobs,
and read live output right in your browser. Start with the forecast-off; the other two go deeper on a single driver.
Your job this week
Finish Project 2 — the credit-default classifier
Due Wednesday, July 15, 12:00 PM (noon) MT. Week 9 is the Project 2 work week — the same
hold-out-and-judge discipline the forecast-off drills below. Open the Project 2 assignment.
Project 2 also lives in its own Canvas module if you want the full brief and data:
Project 2 module.
Forecasting is an FP&A job before it is a stats job.
Every budget, every covenant test, every cash plan starts with a number somebody forecast. The controller's skill
is not one clever model. It is knowing the menu, starting simple, proving a method earns its complexity by beating a
plain SWAG on months it has never seen, and then hedging the number from measured error before it goes in the board deck.
The one idea to leave with — complexity is not accuracy. In the forecast-off, the most complex model only
barely beats a budget-desk method a controller can run in a spreadsheet. Pick the simplest method that
beats your gut, then hedge it with the error you actually measured.
Week 9 Interactives
Start with the forecast-off — it is the whole week on one page. The two driver trainers then go deep on
a single structural idea each: forecasting from a driver you control, and forecasting revenue that has a memory.
-
1
The Forecast-Off Start here
The whole week in one page. Put your gut forecast down first, blind, then watch eight
methods — naive, prior-year, run-rate, moving average, exponential smoothing, linear trend, seasonal naive, and a
fitted SARIMAX — backtest against the same six held-out months of Summit Gear revenue. Read the leaderboard,
turn the knobs, then hedge the number you would report.
Why first: it is the wrap. Every method on the menu, ranked by the miss, with the
SWAG baseline and the hedge step all in one place.
-
2
The marketing driver Driver you control
Forecast sales from a driver you set: marketing spend. Fit sales on spend (the Week 8 OLS with
one predictor), then move next month's planned spend and read the forecast off the line. Push the slider past the
spend you have ever observed and watch the extrapolation danger zone light up — the straight line makes no
promise out there. Plus the causal caveat: a fitted association is not proof spend causes sales.
Why it matters: the cleanest forecast is one you drive from an input you can plan,
as long as you do not extrapolate it past what you have seen.
-
3
Cohort retention Revenue with a memory
Forecast revenue that decays. Acquire a cohort of customers, lose a share of them each year, and
watch this period's acquisitions ripple forward as
(1 - attrition)^t. Shift the acquisition mix toward
referral (low attrition) versus paid ads (high attrition) and watch the long-run revenue base diverge even at
equal acquisition volume. The acquisition mix, not the volume, decides how much of that revenue persists.
Why it matters: subscription and repeat-purchase revenue is not a fresh number each
period — it is what last period's customers are still worth. The channel you acquire on sets that.
The rest of the week
The lecture is the map for the whole survey. Then it is heads-down on Project 2.
How to use this page. Run the forecast-off first for the big picture, then the two driver trainers if you want
to go deeper on spend-driven and cohort-driven forecasts. Keep the AI tutor open the whole time. There is no Lab or
Homework to submit this week — the deliverable in play is Project 2, due July 15 at noon.
Week 9 of ACCTG 5150, Accounting Analytics. The trainers compute everything in your browser
(nothing leaves your machine).