Forecasting Lingo - Terminology Used in Time Series Forecasting for Non-Technical Users

Introduction

So you can better understand how businesses predict future trends. Whether it’s sales, stock prices, or the weather.

What is Time Series Forecasting?

At its most basic, time series forecasting is about using past data (like sales figures or temperatures) to predict future data points. The key is that the data is time-based, meaning each data point is recorded at regular intervals (like daily, weekly, or monthly).

For example, if you have monthly sales figures for the past year, time series forecasting helps you predict what the sales might be for next month.

Common Terms You’ll Hear in Time Series Forecasting

Let's break down the terminology you'll come across in the world of time series forecasting.

1. Model

A model is like a recipe that helps make predictions. It’s a mathematical formula or method that takes your past data and tries to predict future values. Think of it like this: if you bake a cake using a recipe (the model), your ingredients (the past data) will help determine the outcome (the forecast).

2. Variables

In forecasting, variables are the pieces of information that go into the model. For example, if you're forecasting sales, your variables might be past sales, marketing spend, and the time of year. Variables help the model "see" the factors that influence the data you're trying to predict.

There are two types of variables:

• Dependent Variable: The thing you're trying to predict (e.g., sales).
• Independent Variables: These are also known as regressors. They influence the dependent variable. For instance, marketing spend or seasonality can impact your sales forecast.

3. Univariate vs. Multivariate Forecasting

These terms describe the complexity of your forecast:

• Univariate forecasting means you’re only working with one variable, like sales figures over time.
• Multivariate forecasting uses more than one variable. For example, you might forecast sales using not just past sales, but also marketing spend, weather, and time of year.

4. Features and Feature Engineering

Features are the inputs (variables) the model uses to make predictions. Feature engineering is the process of improving these inputs. For example, you might take a basic sales figure and add a new feature that marks whether the day is a holiday, since holidays could affect sales. By tweaking or adding features, you can make your model smarter.

Types of Forecasting Models

Now that we’ve covered basic terms, let's look at the most common forecasting models. You don’t need to understand the math behind these, but knowing what they are can help you ask the right questions.

1. ARIMA

The ARIMA (AutoRegressive Integrated Moving Average) model is a classic choice for forecasting based on past data. It’s great for univariate forecasting and works well if your data shows a clear trend over time. There are pros and cons though.

2. ETS (Exponential Smoothing)

ETS is good for predicting data that has seasonality (patterns that repeat over time, like holiday shopping spikes). It uses past trends and adjusts predictions based on how much the data has changed in the past.

3. Prophet

Prophet is a tool developed by Facebook (now Meta) that is specifically designed for time series forecasting. It automatically detects trends (like a steady increase in sales) and seasonality (like a holiday rush). Prophet is easy to use and helps businesses forecast without needing to know all the technical details of more complicated models.

• Why Use Prophet? It’s great for capturing trends and seasonal patterns, especially for businesses with strong cycles, like retail. Prophet lets you add important external factors (like holiday periods or marketing campaigns) to make more accurate predictions.

4. AutoML (Automated Machine Learning)

AutoML tools can make forecasting easier by automatically selecting and tuning models based on your data. TWhat is Rolling Forecasting?

Rolling forecasting is an ongoing process that regularly updates forecasts based on the latest actuals. Unlike traditional annual budgets, rolling forecasts provide a continuous planning horizon, typically updated quarterly or monthly. This approach enables businesses to stay aligned with current market conditions and adjust their strategies promptly.

Why Switch to Rolling Forecasts?

1. Agility in Decision Making: With rolling forecasts, you can quickly adapt to market changes. This flexibility is crucial in today’s fast-paced business environment.
2. Improved Accuracy: By frequently updating your forecasts, you can reduce the uncertainty and improve the accuracy of your financial planning.
3. Enhanced Visibility: Rolling forecasts provide better visibility into future performance, helping you identify potential gaps and opportunities early on.
4. Strategic Focus: Instead of getting bogged down in the minutiae of annual budgeting, rolling forecasts allow you to focus on key business drivers and strategic initiatives.

Tools and Techniques for Effective Forecasting

To make the most out of rolling forecasts, leveraging advanced forecasting methods and tools is essential. Here’s a breakdown of some powerful techniques and models:

Regressions

Regression analysis is a statistical method used to understand the relationship between variables. It helps in predicting the future values of a dependent variable based on the changes in independent variables.

• Linear Regression: Simple but effective, it’s used when the relationship between variables is linear.
• Multiple Regression: Expands on linear regression by using multiple independent variables to predict the dependent variable.

ARIMA (AutoRegressive Integrated Moving Average)

ARIMA models are used for time series forecasting. They are particularly useful when data shows evidence of non-stationarity, meaning statistical properties change over time.

• AR: AutoRegressive part indicates that the evolving variable of interest is regressed on its own lagged values.
• I: Integrated part indicates differencing of raw observations to make the time series stationary.
• MA: Moving Average part indicates that the regression error is actually a linear combination of error terms whose values occurred contemporaneously and at various times in the past.

SARIMA (Seasonal ARIMA)

SARIMA extends ARIMA by adding a seasonal component to the model. This is useful for data with regular, repeating patterns.

• S: Seasonal component helps in adjusting the model to account for seasonality in the data.

SARIMAX (Seasonal ARIMA with eXogenous variables)

SARIMAX goes a step further by incorporating external variables (exogenous variables) into the SARIMA model. This is particularly useful when other factors outside the time series influence the data.

Prophet

Prophet is an open-source forecasting tool developed by Facebook. It’s designed to handle time series data that have strong seasonal effects and several seasons of historical data.

• Ease of Use: Prophet is user-friendly and requires minimal data preprocessing.
• Flexibility: It can handle missing data and shifts in the trend.

MLForecast

MLForecast is a machine learning-based forecasting tool that leverages advanced algorithms to predict future trends.

• Advanced Models: Uses models like Gradient Boosting Machines (GBM) and Neural Networks.
• Feature Engineering: Automatically creates relevant features from the time series data.

NeuralProphet

NeuralProphet combines the best features of Prophet and neural networks to provide high accuracy in forecasting.

• Hybrid Approach: Uses both statistical and machine learning techniques.
• High Accuracy: Particularly effective for complex time series data.

Implementing Rolling Forecasts: Step-by-Step Guide

Step 1: Use Dedicated Applications

Spreadsheets might be great for quick calculations, but they’re not cut out for rolling forecasts. A dedicated cloud-based application can streamline the process, reducing errors and saving time.

• Benefits: Automation, error reduction, real-time data integration.
• Examples: Trackura.

Step 2: Focus on Key Business Drivers

Don’t get bogged down in details. Focus your forecasts on the significant drivers of your business, such as revenue, costs, and capital expenditures.

• Identify Drivers: Determine which variables have the most impact on your business.
• High-Level Focus: Keep the forecasts at a high level to avoid getting lost in the weeds.

Step 3: Model Multiple Scenarios

Use rolling forecasts to explore various “what-if” scenarios. This helps you prepare for different potential outcomes and make informed decisions.

• Scenario Planning: Create best-case, worst-case, and most-likely scenarios.
• Flexibility: Quickly adjust your plans based on these scenarios.

Step 4: Separate Forecasts from Targets

Ensure that rolling forecasts are used as a management tool rather than an evaluation tool. This encourages honest and accurate forecasting.

• Avoid Linking to Rewards: Don’t tie forecasts to performance rewards to prevent number padding.
• Transparent Process: Keep the forecasting process open and collaborative.

Step 5: Choose the Right Forecasting Horizon

The appropriate forecasting horizon depends on your industry and business needs. Typically, forecasting four to eight quarters ahead is recommended.

• Industry Specifics: Adjust the horizon based on how quickly your industry changes.
• Review and Adjust: Regularly review the forecasting horizon to ensure it remains relevant.

Practical Tips for Better Forecasting

1. Regular Updates: Update your forecasts regularly to reflect the latest data and market conditions.
2. Collaboration: Involve different departments in the forecasting process to get a comprehensive view.
3. Technology Integration: Integrate your forecasting tool with other business systems for seamless data flow.
4. Continuous Improvement: Continuously refine your forecasting models and techniques based on past performance.

Common Pitfalls to Avoid

1. Over-Reliance on Historical Data: While historical data is useful, it’s not always indicative of future trends. Combine it with real-time data and market analysis.
2. Ignoring External Factors: Consider external factors such as economic conditions, competitor actions, and regulatory changes in your forecasts.
3. Lack of Flexibility: Be prepared to adjust your forecasts as new information becomes available.
4. Inadequate Communication: Ensure that the insights from rolling forecasts are communicated effectively to all stakeholders.

To Sum It Up...

Rolling forecasts are a game-changer for businesses looking to stay ahead in a dynamic market. By leveraging advanced forecasting tools and techniques like regressions, ARIMA, SARIMA, SARIMAX, Prophet, MLForecast, and NeuralProphet, you can enhance the accuracy and agility of your financial planning. Remember, the goal is not just to predict the future but to be prepared for it. So, set your course with rolling forecasts, and navigate your business towards success.

A Friendly Reminder

Forecasting might sound like a daunting task, but with the right tools and approach, it can be as smooth as a leisurely sail on a sunny day. And remember, even the best forecasts won’t always be 100% accurate—just like weather forecasts. But they’ll definitely help you avoid sailing blindly into a hurricane. Happy forecasting!By following these steps and leveraging the right tools, you can transform your forecasting process and ensure that your business is always ready to navigate the future, no matter what storms may come your way.hese tools test multiple algorithms and pick the one that works best for your specific data set, taking out the guesswork.

Popular AutoML tools for time series forecasting include:

• AutoGluon: A tool from Amazon that’s easy to use, even for non-technical users.
• Nixtla: This tool offers a mix of machine learning and statistical methods for time series forecasting.

Important Concepts in Forecasting

1. Seasonality

Seasonality refers to patterns in data that repeat over regular intervals. For example, a retail business might see a big sales jump every December during the holiday season. Forecasting models like ETS and Prophet are great for detecting and predicting these kinds of patterns.

2. Lag Variables

A lag variable is data from previous time periods used to predict future outcomes. For example, last month’s sales might be used to predict next month’s sales. By including lag variables, the model can learn how past events affect future outcomes.

3. Overfitting

Overfitting happens when a model is too good at predicting past data but struggles to predict future data. It’s like memorizing answers for a test rather than understanding the concepts. Overfitted models might make poor predictions because they’re "too specific" to the past data.

Key Takeaways for Non-Technical Users

• Start Simple: If you’re new to forecasting, start with simpler models like ARIMA or ETS for univariate data.
• Multivariate Forecasting: If your forecast depends on more than one factor (like sales and marketing spend), you’ll need a multivariate model.
• AutoML and Prophet: Tools like Prophet and AutoML solutions make forecasting accessible, even if you don’t have a deep technical background.
• Feature Engineering is Important: Pay attention to what variables (features) you feed into the model. Tweaking these can greatly improve your forecasts.

With a basic understanding of these concepts, you'll be ready to dive deeper into forecasting and make sense of predictions that drive business decisions. Whether you're working with sales data or planning for the future, these terms will give you a solid foundation in time series forecasting.

This guide should give you the confidence to understand the terminology of time series forecasting, whether you're using it yourself or working with someone to create forecasts for your business. By knowing the basics, you can make smarter decisions and ask the right questions about future predictions. If you're looking for forecasting software that's usable by regular folks, check out what Trackura can do for you.