Compass Forecasting Logo

Statistical Forecasting

Understanding the Statistical Forecast Method

When managing your worksheet accounts, you have the option to select Statistical Forecast. This is the most advanced forecasting method available in CompassForecasting. Instead of relying on a simple average or a manual guess, it uses a mathematical algorithm to analyze your historical data and predict the future.

We use an industry-standard forecasting algorithm called the Holt-Winters method. This guide explains how it works in plain English and how you can tweak its settings to get the perfect forecast.

What is Holt-Winters?

Imagine you run an ice cream shop. Your sales probably do three things:

  1. Level: You have a baseline amount of sales you make on an average day.
  2. Trend: Over the last few years, your business has been growing, so your baseline sales are slowly creeping up.
  3. Seasonality: You sell way more ice cream in the summer than in the winter.

The Holt-Winters method is a mathematical formula that looks at your past data, separates it into those three exact buckets (Level, Trend, and Seasonality), and projects them into the future. It recognizes that last summer was busy, notes that your business is growing 5% year-over-year, and calculates exactly what next summer should look like based on those patterns.

The Settings Explained

When you select Statistical Forecast, you can adjust several settings to control how the algorithm thinks.

1. Historical Periods (e.g., 36 Months)

This tells the algorithm how far back in time it should look to learn about your business.

  • Default: Usually set to 36 periods (3 years).
  • When to change it: If your business completely pivoted or changed dramatically 2 years ago, data from 3 years ago might confuse the algorithm. In that case, you might lower this number to 24 so it only looks at the “new” version of your business.

2. Adj (Adjustment Percentage)

The algorithm predicts what will happen if things continue exactly as they have in the past. But what if you know something the algorithm doesn’t?

  • What it does: Allows you to manually bump the entire mathematical forecast up or down by a specific percentage.
  • Example: The algorithm predicts $100,000 in sales next year. However, you know you just hired two new salespeople. You can put 10 in the Adj box to automatically boost the statistical forecast by +10%.

3. Seasonality

This tells the algorithm how long your business’s “cycle” is before it repeats.

  • Auto: The system will try to figure it out automatically.
  • Monthly (12): Tells the system that your cycle repeats every 12 months (e.g., December is always busy, January is always quiet). This is the most common setting for standard businesses.
  • Quarterly (4): Use this if your cycle repeats exactly every 4 quarters.
  • None (1): Tells the system you have absolutely no seasonal cycle; your sales are flat year-round.

4. Seasonal Model (Additive vs. Multiplicative)

This is an advanced setting that tells the algorithm how your seasonal spikes behave as your business grows.

  • Additive: Use this if your seasonal spikes stay the same dollar amount even as you grow. (Example: Every December, you make exactly $10,000 more than usual, regardless of how big your company gets).
  • Multiplicative: Use this if your seasonal spikes grow proportionally as your business grows. (Example: Every December, your sales increase by exactly 50%. If your base sales double over 5 years, the size of that December spike will also double).

What is RMSE?

When you review a Statistical Forecast chart, you will see a metric called RMSE (Root Mean Square Error).

While it sounds intimidating, RMSE is just a grade for how accurate the algorithm is.

In plain English: RMSE is the average dollar amount the algorithm was “wrong by” when it tried to guess your historical data.

  • If your RMSE is $500, it means that on average, the algorithm’s predictions miss your actual real-world numbers by about $500 per month.
  • Lower is always better. A lower RMSE means the algorithm’s line tightly hugs your historical data, meaning the future forecast is highly reliable.
  • Higher means less reliable. If the RMSE is huge compared to your total revenue, it means your historical data is so random and chaotic that the algorithm is struggling to find a clear pattern. In this case, you might want to switch to a simpler method like a 12-Month Average.