Estimated reading time: 10 minutes
As we live in an increasingly data-driven world, it has become increasingly important for real-estate investors to be able to accurately predict the future worth of properties in a suburb. More and more algorithms are being developed which can forecast house prices. But have you ever wondered how accurate these house price forecasts are?
HtAG Analytics uses house price forecasts to benchmark the capital growth potential of property markets. Our algorithm is specifically designed to identify long-term trends in house prices for a particular suburb or LGA. This makes our forecasts ideal for those looking for long-term investments, as many investment grade markets can deliver remarkable capital growth returns over extended periods of time (e.g. 7+ years).
In this article we will examine what our model was predicting a for year 2022 at the end of 2021 via a forecast backtest.
Table of Contents
House Prices Are Not What You Think
Before we look at backtesting the forecasts, there are several concepts you will need to understand about the housing market data in general.
Analysis of house price data for suburban property markets across the country reveals that median prices are an unreliable measure of house prices in suburban areas.
Asevident on the chart below, the median price data shows considerable fluctuation month-on-month, making it difficult to identify any clear trends. To address this issue, HtAG Analytics developed a proprietary metric – Typical Price – to provide a more accurate understanding of suburb house prices.
The chart above presents house sales and prices for the Suburb of Krabar, NSW. The black dots represent actual house sales, while the faint blue line shows the monthly median prices based on these sales.
The dark blue line is the monthly typical price, which we will be back-testing.
Unlike the median price, the typical price is a dynamically revised metric that is established through the process of data fitting. This methodology is used to generate prices for the entire historic data range and is recalculated every month, including for previous periods.
In a way, typical price predicts both the current and historical prices in a suburb using all data available. Despite its revisionary nature, it is a much more accurate metric than median prices at the suburb and LGA levels.
How Do We Backtest the Typical Price Forecast?
We simply withhold 1 year worth of data from the model and execute our data pipeline. The process runs in 3 steps:
- Generate historical typical price (orange line)
- Generate the forecast (dashed orange line)
- Estimate the error rate by comparing the actual data with the forecast (dark blue line vs the dashed orange line)
You will notice that the backtest typical price (orange line) may vary slightly from actual typical price (dark blue line) on the chart above. This is due to the idiosyncrasies of the data fit process we explained earlier. For comparison, despite the slight variance, this metric is still much more accurate than the median price (faint blue line).
Our team runs quarterly backtests to determine how closely price changes align with the estimated trend. The error rate metric in our market reports captures the results of these backtests – the lower the error, the better the market aligns with the projected trend.
Is the Error Rate More Important than the House Price Forecast?
Error rate is a critical metric used to evaluate the accuracy of a forecast. It is calculated by comparing the actual house prices to the back-dated trend forecast as a percentage and provides several valuable insights. This comparison allows for performance feedback, which can be used to provide further context for future predictions.
The purpose of HtAG’s forecast algorithm is to project the overall trajectory of house price growth. It is not meant to provide precise figures, but rather a long-term trend expressed as a percentage. In other words, the dashed orange line on the graph is not designed to forecast the actual typical price within the forecast period.
So why do we do it?
Projected trend allows us to calculate likely percentage growth for future prices in any given market. And by comparing the projected trend to the actual price change within the short-term backtest period, we can calculate the degree that the market fluctuates around the estimated long-term trend. Repeatedly re-generating and measuring both the trend and the error helps us establish “the corridor” for future house price changes.
The chart above shows the results for 24 forecast backtests over a 2 year period with a cumulative error of 5.12%. The cumulative error allows us to establish the range of future annual price change in this market.
HtAG’s Capital Growth (CG) metric is a combination of house price trend estimate and cumulative error rate calculated via regular backtesting. The trend is calculated monthly using the latest sales data. The error is calculated every quarter and added to the cumulative error rate:
- CG Low = Estimated house price trend – Cumulative error
- CG High = Estimated house price trend + Cumulative error
A higher error rate indicates a greater degree of uncertainty and suggests that the possible range of price growth can vary widely. On the other hand, a lower error rate suggests that price growth is more likely to stay within tighter boundaries. Knowing the trend and the error rate can help investors make more informed decisions when comparing property markets.
Piecing It All Together via the CG and RCS™ Metrics
The CG (Capital Growth) Low/High range provides an excellent opportunity to examine potential worst and best-case scenarios for annual price growth in over 5,000 property markets included in our reports. It is recommended that this metric is utilised as a range and no attempt is made to discern an average value.
When considering a market’s capital growth potential, other factors like the supply/demand balance and the overall state of the market should not be overlooked. To gain further insight into all the metrics that influence capital growth, visit our data dictionary.
HtAG’s RCS™ (Relative Composite Score) metric encompasses both the projected price trend and cumulative error. Additionally, the RCS Methodology looks at other crucial market fundamentals and supply/demand metrics.
We recommend that the RCS™ Capital Growth metric is used as a guide when comparing housing markets. Generally speaking, if the RCS™ Capital Growth metric is high, then it is more likely that the high range of (CG) Capital Growth will occur in future years.
The Latest Backtest Winners and Losers
Our dataset contains data for more than 5,000 suburbs. To illustrate the internal processes, we present the backtesting results from a sample of 9 property markets.
According to the error rate in the backtest, we can classify winners, contenders and losers:
- Winners have an error rate below 3%,
- contenders fall between 3-7%,
- while losers are above 7%.
December 2022 typical price forecast backtest winners are:
Dean Park, NSW: 0.82% error
The Ponds, NSW: 1.11% error
Jerrabomberra, NSW: 1.10% error
December 2022 typical price forecast backtest contenders are:
Emerton, NSW: 4.54% error
Bundall, QLD 5.28% error
Yakamia, WA: 5.75% error
December 2022 typical price forecast backtest losers are:
Albany, WA: 12.18% error
Worongary, QLD: 13.8% error
Milang, SA 16.31% error
It’s important to keep in mind that the error rate serves as an indicator to measure the deviation of price change from the long-term trend. Therefore, you shouldn’t be too encouraged by low error rates or discouraged by higher values. This metric acts as an excellent predictor for the volatility of house prices in any given housing market.
The median error rate across the December 2022 backtest was 6.17%. The state-by-state breakdown of results is detailed below. During this period, it is evident that QLD, SA, and TAS experienced the greatest deviation from the long-term house price trend.
| State/Territory | Median Error % |
|---|---|
| ACT | 4.05 |
| NSW | 5.23 |
| NT | 4.41 |
| QLD | 9.87 |
| SA | 9.48 |
| TAS | 9.02 |
| VIC | 4.09 |
| WA | 4.75 |
| Overall | 6.17 |
As mentioned before, the backtest error data is collected and averaged in a cumulative pool that helps determine the long-term range for future price growth/decline. To view the cumulative error rate for any property market in our reports:
- Head over to the ranking table on the home page
- Click on Expert View toggle below the table
- Scroll to the right to see the error rate column
Summary
Understanding the trend and error rate generated via a backtesting process can help investors make more informed decisions when comparing property markets.
- HtAG Analytics developed a proprietary metric – Typical Price – to provide an accurate understanding of house prices.
- Our algorithm is designed to identify long-term trends in typical price i.e. the forecast is calculated based on full history of house price movements in a property market and is an estimate of house price trajectory for the next 7+ years.
- The backtesting process involves withholding 1 year worth of data to compare forecasted and actual typical price.
- Error rate is a critical metric used to establish price volatility and future range of house price changes, calculated during the backtesting process.
- Lower error rates suggest that prices will stay within tighter boundaries, while higher rates suggest a wider range of potential growth/decline.
- By combining forecast trend and cumulative error, a low and high range of future house price growth can be established.
Establishing house price trend and backtest error rate are a part of a larger landscape. These 2 metrics serve as input parameters into the RCS™ Capital Growth and RCS™ Lower Risk metrics.
We recommend relying on RCS™ for decision making when selecting the best investment locations to create or expand you property investment portfolio, whilst using the CG High/Low metric to establish the range of future house price change.