Welcome to Your Journey into Predictive Analytics!
Hello there! If you are preparing for Exam PA, you are already well on your way to mastering one of the most exciting fields in actuarial science. In this chapter, we are going to look at the three main pillars of analytics: Descriptive, Predictive, and Prescriptive.
Think of this as the "map" of your project. Before you start building complex models or writing R code, you need to know what kind of question you are trying to answer. Are you looking at the past, guessing the future, or deciding on a plan? Let’s dive in!
1. Descriptive Analytics: "What Happened?"
Descriptive analytics is the most basic form of analysis. It looks at historical data to summarize what has already occurred. In the actuarial world, we do this all the time when we look at last year's claims or current policyholder demographics.
The Goal: To provide a clear picture of the past and identify patterns or trends that have already taken place.
Common Tools:
- Summary Statistics: Calculating the mean \( \bar{x} \), median, and standard deviation \( \sigma \).
- Data Visualization: Using histograms, bar charts, and scatterplots to "see" the data.
- Reporting: Creating dashboards that show how many policies were sold last month.
Real-World Analogy: Imagine you are checking your bank statement at the end of the month. You are seeing where your money went, which categories you spent the most on, and what your final balance is. You aren't changing anything; you're just observing the facts.
Quick Review: Descriptive Analytics
Key Question: What happened?
Focus: Hindsight and historical data.
Example: "Last year, our average claim cost was \$5,000."
Don’t worry if this seems simple! Every great predictive model starts with a strong descriptive foundation. You have to understand the past before you can predict the future.
2. Predictive Analytics: "What Might Happen?"
This is the heart of Exam PA! Predictive analytics uses historical data and statistical modeling to identify the likelihood of future outcomes. We aren't 100% certain what will happen, but we use math to make the most educated guess possible.
The Goal: To find relationships between variables and use them to forecast future events or behaviors.
Common Tools:
- Generalized Linear Models (GLMs): Predicting a target variable like claim frequency.
- Decision Trees: Grouping policyholders based on their risk factors.
- Time Series: Forecasting future values based on past trends.
Real-World Analogy: Think of a weather forecast. Meteorologists look at current pressure, temperature, and wind patterns to say, "There is an 80% chance of rain tomorrow." They aren't certain it will rain, but the data suggests it is highly likely.
Did you know? Predictive analytics doesn't just predict the "future" in terms of time; it can also predict "unknown" values. For example, predicting whether a current customer will renew their policy next month is a predictive task.
Quick Review: Predictive Analytics
Key Question: What is likely to happen?
Focus: Insight and forecasting.
Example: "Based on their driving record, this customer has a 15% chance of filing a claim next year."
3. Prescriptive Analytics: "What Should We Do?"
Prescriptive analytics is the "final frontier." It goes beyond just predicting what will happen and actually recommends a specific course of action to reach a goal. It answers the question: "How can we make the best outcome happen?"
The Goal: To optimize decision-making by considering multiple possible futures and recommending the best path forward.
Common Tools:
- Optimization: Finding the "best" price to maximize profit without losing too many customers.
- Simulation: Running thousands of "What if?" scenarios to see which strategy performs best under stress.
- Decision Rules: Automated systems that say, "If the predicted risk is > X, then do Y."
Real-World Analogy: Think of a GPS (Google Maps). Descriptive is showing where you are. Predictive is saying "There is traffic ahead, so you will arrive in 40 minutes." Prescriptive is saying "Take the next right turn to save 10 minutes." It gives you a recommendation to achieve your goal.
Quick Review: Prescriptive Analytics
Key Question: What should we do?
Focus: Optimization and strategy.
Example: "To maximize our total profit, we should increase premiums for high-risk drivers by 10% and offer a 5% discount to low-risk drivers."
Summary: The Analytics Spectrum
It is helpful to think of these three types as a progression. As you move from Descriptive to Prescriptive, the complexity increases, but so does the value added to the business.
1. Descriptive: Low complexity, low value (hindsight).
2. Predictive: Medium complexity, medium value (insight).
3. Prescriptive: High complexity, high value (foresight/optimization).
Memory Aid: The Triple-P Mnemonic
- Descriptive = Data (What do we have?)
- Predictive = Probability (What might happen?)
- Prescriptive = Plan (What should we do?)
Common Mistakes to Avoid
1. Confusing Prediction with Prescription: A common error on the exam is stopping at a prediction. If the business problem asks how to reduce costs, simply predicting who will have high costs (Predictive) isn't enough. You must recommend an action (Prescriptive), such as a wellness program or a price change.
2. Ignoring Descriptive Analysis: Students often want to jump straight into the fancy models. However, without descriptive analytics, you might miss errors in your data or fail to realize that your "target" variable is skewed.
Key Takeaway for Exam PA
On the exam, always read the Business Problem carefully.
- If they want to know patterns, use Descriptive tools.
- If they want to estimate a future value, use Predictive tools.
- If they want to make a decision or optimize a result, use Prescriptive logic.
You've got this! Understanding these definitions is your first step toward defining the problem correctly, which is worth significant points on the actual exam. Keep moving forward!