Baseline Trend Forecasting

Why this matters

Marketing Analysts often need a quick, reliable forecast to set targets and budgets. Baseline trend forecasting gives a defensible starting point for planning campaigns, inventory, and workload.

• Set next month’s leads target from a steady growth trend.
• Create a budget baseline for ad spend based on historical growth.
• Estimate support volume next quarter to staff teams appropriately.

Who this is for

• Marketing Analysts and coordinators who report metrics weekly or monthly.
• People new to forecasting who need a strong baseline before advanced models.

Prerequisites

• Comfort with averages and linear equations.
• Basic spreadsheet skills (SUM, AVERAGE, TREND/LINEST or simple regression).

Concept explained simply

Baseline trend forecasting predicts the next value by assuming the recent trend continues. You can do this with:

• Naive: next value equals the last observed value.
• Moving Average (MA): next value equals the average of the last k values.
• Linear Trend (OLS): fit a straight line through time; extend it forward.
• Holt’s Linear (Exponential Smoothing with trend): a smooth level plus a smooth trend.

Mental model

Think of your metric as a straight road with some bumps. Naive takes the last point on the road. Moving average smooths bumps. Linear trend draws the road’s slope. Holt’s keeps the slope but adapts faster to changes.

When to use which method

• Naive: very short series or highly random data.
• MA: noisy but roughly stable trend, needs smoothing.
• Linear Trend: clear steady growth or decline.
• Holt’s: trend exists and you want responsiveness without overreacting.

Worked examples

Example 1 — Sessions: naive, 3-MA, linear trend

Weekly website sessions (10 weeks): 940, 960, 970, 995, 1005, 1010, 1030, 1040, 1055, 1070.

• Naive forecast (week 11): 1070.
• 3-week MA (week 11): (1040 + 1055 + 1070) / 3 = 1055.
• Linear trend (OLS): y ≈ 931 + 13.9·t. At t = 11 → 931 + 13.9×11 ≈ 1084.

How the linear trend was obtained

Fit y = a + b·t using ordinary least squares on t = 1..10. The slope is about 13.9 and intercept about 931. Any spreadsheet can compute this via TREND/LINEST or a simple regression.

Example 2 — Holt’s linear forecast

Given last estimated level L = 1065 and trend T = 14 (after time 10), the 1-step ahead forecast for time 11 is F11 = L + T = 1079. Holt’s method updates L and T every period using smoothing parameters α (level) and β (trend) between 0 and 1.

Update equations (in words)

• New level = blend of actual and (previous level + trend).
• New trend = blend of (new level − previous level) and previous trend.
• Forecast next = new level + new trend.

Example 3 — Pick a baseline with backtesting

For the last 12 months of leads, test Naive vs Linear Trend using a rolling 1-step-ahead backtest on the last 4 months. Compute MAE (mean absolute error) and choose the lower. If Linear Trend has MAE 18 vs Naive 26, pick Linear Trend as your baseline. Keep the other method as a sanity check.

Step-by-step process

1. Plot your data (even a quick sparkline helps spot trend/seasonality).
2. Pick a baseline method (start with Linear Trend or MA; keep Naive as a reference).
3. Fit the method on the historical period you’ll use.
4. Backtest on a small holdout (e.g., last 4 periods); compute MAE or MAPE.
5. Choose the simplest method that performs well and document the choice.
6. Forecast the next period(s) and round appropriately for the metric.

Common mistakes and self-check

• Ignoring clear seasonality. Self-check: Compare same weekday/month vs overall trend; if patterns recur, note it and consider separate baselines by season later.
• Over-smoothing with a long MA window. Self-check: Does the forecast lag turning points badly? Try a shorter window.
• Using R-squared alone. Self-check: Prefer MAE/MAPE on a holdout; they reflect forecast error.
• Forecasting far beyond the data. Self-check: Limit baseline trend horizons; uncertainty grows quickly.
• Not documenting assumptions. Self-check: Write 1–2 lines: method, window/parameters, holdout error.

Exercises

Do these in a spreadsheet or calculator. Answers are available in collapsible sections. After finishing, use the checklist below.

Exercise 1 — Compare naive, 3-MA, and linear trend

Data: Weekly sessions for weeks 1–10 = 940, 960, 970, 995, 1005, 1010, 1030, 1040, 1055, 1070.

1. Compute the week 11 forecast using: (a) Naive, (b) 3-week Moving Average, (c) Linear Trend (fit y = a + b·t, t=1..10).
2. Which method gives the highest forecast, and why?

Show solution

Naive: 1070. 3-MA: (1040+1055+1070)/3 = 1055. Linear Trend: approximately y = 931 + 13.9·t → at t=11: ≈ 1084. Highest is Linear Trend due to upward slope.

Exercise 2 — One-step Holt update

Given L9 = 1048, T9 = 13, α = 0.3, β = 0.2, and y10 = 1070.

1. Compute L10 = α·y10 + (1−α)·(L9 + T9).
2. Compute T10 = β·(L10 − L9) + (1−β)·T9.
3. Compute F11 = L10 + T10.

Show solution

L10 = 0.3·1070 + 0.7·(1048+13) = 321 + 742.7 = 1063.7.
T10 = 0.2·(1063.7−1048) + 0.8·13 = 3.14 + 10.4 = 13.54.
F11 = 1063.7 + 13.54 = 1077.24.

Exercise checklist

• [ ] I can compute 1-step ahead Naive, MA, and Linear Trend forecasts.
• [ ] I can perform a Holt one-step update and produce a forecast.
• [ ] I used a small holdout to compare MAE across methods.
• [ ] I wrote down method, parameters, and error metric.

Practical projects

• Baseline pipeline: Build a small sheet that takes the last 12 data points and outputs Naive, 3-MA (k=3), Linear Trend, and Holt forecasts plus MAE on the last 3 points.
• Quarterly target setting: Use Linear Trend to propose next quarter’s leads target; include an error band based on recent MAE.
• Stakeholder one-pager: Visualize last 12 points, chosen baseline forecast, and a short note on assumptions.

Learning path

1. Start with Naive and Moving Average for quick baselines.
2. Add Linear Trend (OLS) for steady growth/decline.
3. Adopt Holt’s Linear for adaptive trend when data updates frequently.
4. Later, incorporate seasonality handling (beyond this subskill).

Next steps

• Backtest two methods on your real metric and choose one baseline.
• Document assumptions and share with your team for feedback.
• Move on to handling seasonality once your baseline trend is stable.

Mini challenge

Take your last 8 weeks of a metric you track. Compute Naive, 3-MA (k=3), and Linear Trend forecasts for the next week. Pick one as your baseline, and justify your choice in 2 sentences using holdout MAE.

Quick test

Anyone can take the test. Only logged-in users will have their progress saved.