Last week, my client Michael excitedly sent me a link: 'Ah Ken, take a look at this AI tool, it says that Taikoo Shing will drop by 3% next month. Should I wait a bit before entering the market?' I opened it and found yet another AI platform claiming to 'accurately predict property prices.' This was already the fifth client this month asking me a similar question.
With the sudden rise of ChatGPT and various AI tools, a lot of 'AI real estate price prediction' services have emerged in the Hong Kong property market. Some paid platforms even claim an accuracy rate of up to 85%, attracting many prospective buyers and investors to subscribe. But as a veteran with 15 years of experience in the real estate industry, I have to tell you: predicting property prices has never been a simple mathematical formula, and AI is not a crystal ball.
In today's article, I will use the most straightforward approach to break down the real capabilities and limitations of AI in predicting housing prices, as well as the investment indicators you should truly pay attention to. Whether you are a first-time buyer preparing to get on the property ladder, or an investor looking to leverage for cashing out, after reading this, you will understand: blindly trusting AI predictions can easily make you miss opportunities, or even buy at a high point.
:::tip Expert Tips AI can be an auxiliary tool, but it can never replace professional judgment. True real estate investment decisions require a combination of market data, policy trends, personal financial situation, and on-site inspections. :::
The Principles of AI Predicting Property Prices: How Does It Actually Work?
How Machine Learning 'Learns' Real Estate Data
Most AI tools for predicting housing prices are based on "machine learning" technology. In simple terms, the system analyzes historical data from the past 5-10 years, including:
- Transaction Price Trends: Centaline City Leading Index (CCL), Midland Property Price Index, etc.
- Macroeconomic Indicators: Unemployment rate, GDP growth, inflation rate
- Mortgage Rate Changes: Prime rate (P), Hibor (H) trends
- Supply Data: Volume of newly launched first-hand properties, number of units under construction
- Policy Factors: Cooling measures, mortgage loan ratios, stamp duty adjustments
AI will try to find the 'correlations' between these data and then build a predictive model. For example, when the unemployment rate rises + mortgage rates increase + new housing supply increases, past housing prices usually fall by 2-4%. The system will then 'learn' this pattern and apply it to future predictions.
Three Major Types of AI Prediction Technologies
The AI real estate price prediction tools currently on the market are mainly divided into three categories:
- Time Series Analysis (Time Series): Predict future trends purely based on past price movements (similar to technical analysis)
- Regression Models (Regression): Analyze the impact of multiple variables (such as interest rates, supply) on housing prices
- Deep Learning (Deep Learning): More complex neural networks that can handle nonlinear relationships and large amounts of data
:::highlight Insider's perspective Most free or low-cost AI forecasting tools actually only use the most basic time series analysis. This method is fairly accurate when trends continue, but once there is a sudden policy change (such as the full removal of restrictions in 2024), the forecasts will be completely off. :::
Why does AI claim an "85% accuracy"?
You might ask, 'Since there is data to support it, why can't we trust it?' The problem lies in the definition of this '85% accuracy rate.'
Most platforms calculate accuracy in the following way: as long as the predicted direction is correct, it counts as accurate. For example, if an AI predicts that housing prices will drop next month and they do drop (whether by 0.5% or 5%), it is considered a 'successful prediction.' But for investors, the magnitude of the drop is what really matters — a 0.5% drop might make you continue holding, while a 5% drop might force you to stop loss and exit.
A bigger problem is that these 'historical accuracies' are all based on backtesting, which means testing the model's performance using past data. But the real estate market is dynamic, and patterns that worked in the past may not apply to the future. Social events in 2019, the pandemic in 2020, and the deregulation in 2024—these 'black swan events' are not included in the AI's training data.
Real Cases: Three Classic Scenarios Where AI Predictions Fell Short
Case 1: On the Eve of Lifting the Ban in February 2024
In January 2024, I had a client, Sarah, who was preparing to buy a property in Tseung Kwan O. At that time, she used an AI prediction tool, and the system indicated, 'Property prices will remain flat for the next three months, it is recommended to wait.' As a result, at the end of February, the government suddenly announced a full withdrawal of the cooling measures, and the property market immediately rebounded. The bargain property she was interested in went up 8% in price within a week.
Reasons for AI Inaccuracy: Sudden policy changes are considered 'exogenous variables' and cannot be predicted from historical data. Even the most advanced deep learning models cannot foresee when the government will adjust policies.
:::warning Guide to Avoiding Pitfalls If you are in a 'policy-sensitive period' (such as when the government frequently releases signals of relaxation), do not overly rely on AI predictions. At this time, policy trends are more important than historical data. :::
Case 2: Misjudgment of the 2023 Interest Rate Hike Cycle
In 2023, the US Federal Reserve raised interest rates consecutively, and Hong Kong mortgage rates surged from 2% to 4.5%. Many AI tools at the time predicted that "housing prices would drop by 15-20%" because historical data showed that "interest rate hikes = housing price declines".
But the actual situation is: housing prices only slightly adjusted by 5-8%, and some high-quality estates (such as Taikoo Shing and Mei Foo) even rose slightly against the market trend. The reason is that new factors have emerged in Hong Kong at the same time, such as the 'return of the emigration wave,' 'increase in mainland buyers,' and 'shortage of supply,' all of which are not included in AI's prediction model.
Reasons for AI Inaccuracy: The real estate market is affected by multiple factors, and a single variable (such as interest rates) cannot determine the trend. AI finds it difficult to capture soft factors such as 'market sentiment' and 'capital flow'.
Case 3: Regional Differences Ignored
My other client, Tommy, wanted to invest in the Tsuen Wan area. AI tools predicted that "housing prices in West New Territories would rise by 3%." So he purchased a unit above Tsuen Wan West Station, but six months later, due to delays in transportation facilities in the area, the housing prices did not rise and instead fell by 2%.
During the same period, the old buildings in Tsuen Wan town center (such as around Tsuen Wan Plaza) rose by 6% due to redevelopment expectations.
Reasons for AI Inaccuracy: Most AI tools can only predict trends at the 'overall area' level and cannot be broken down to the 'housing estate level' or 'street level.' However, real investment opportunities are often hidden in these microscopic differences.
:::tip Experts recommend If you want to use AI to assist in decision-making, remember to 'validate in layers': first look at the overall real estate market trends, then examine regional trends, and finally, be sure to conduct on-site inspections of the specific housing estates' supply and demand conditions, supporting facilities development, and the quality of the owners. :::
The Five Major Limitations of AI Predictions: The Truth You Must Know
Limitation One: Unable to Predict 'Black Swan Events'
The biggest risks in the real estate market often come from unexpected events: sudden policy changes, economic crises, geopolitical issues, outbreaks of epidemics, and so on. These events occur very infrequently in historical data, and AI cannot learn or predict them at all.
The social events of 2019, the pandemic in 2020, the interest rate hikes in 2022, and the easing in 2024 are all 'black swans.' If you blindly trust AI predictions during these critical moments, you may make wrong decisions at any time.
Limitation Two: The Problem of Data Lag
The data that AI relies on mostly has a lag of 1-2 months. For example, government-released transaction data, mortgage approval data, vacancy rate data, etc., are all 'in the past.' But the property market changes very quickly, especially during periods of policy shifts, and a one-month lag is already enough to make forecasts fail.
I have seen many clients who, upon seeing the data 'last month's property prices fell,' assumed the market was deteriorating, and as a result, missed the rebound opportunity of that month.
Limitation Three: Ignoring Local Factors Such as 'Supply, Equilibrium, and Rent'
Hong Kong's property market has a unique phenomenon: mortgage payments are cheaper than rent. When mortgage payments are lower than rental expenses, even if property prices fluctuate in the short term, owner-occupiers will still choose to enter the market. This kind of 'homeownership mentality' is difficult for AI to quantify.
Similarly, Hong Kong people's confidence in 'bricks,' their obsession with 'getting on the property ladder,' and their pursuit of 'prestigious school networks'—these cultural factors all influence the real estate market, but AI cannot understand them.
Limitation Four: Unable to Assess the Quality of Individual Housing Estates
AI can tell you 'the property prices in Sha Tin District are expected to rise by 2%,' but it cannot tell you:
- Which has a higher appreciation, First City or Yu Chui Court?
- Which building has better management quality?
- Which unit orientation retains value the best?
- Are the owners mostly investors (quick turnover) or homeowners (stable holding)?
These 'micro details' are the real key factors affecting investment returns, but they require on-site inspection and professional judgment, which AI cannot do.
Limitation Five: The Longer the Forecast Period, the Greater the Error
Most AI tools can only predict short-term trends for the "next 1-3 months," and their accuracy is already limited. If you want to know the housing prices for the "next year" or even the "next three years," AI's predictions are basically just "guesses."
Property market investment is a long-term strategy. If you only look at short-term forecasts, it is easy to be influenced by market noise and make wrong decisions.
:::warning Risk Warning Never give up entering the market just because AI predicts 'it will fall next month,' and do not blindly chase highs just because it predicts 'it will rise next month.' Property investment should focus on the 3-5 year long-term trend, not short-term fluctuations. :::
How Smart Investors Can Make Good Use of AI: Four Practical Strategies
Although AI predictions have limitations, this does not mean they are completely useless. The key is to understand how to 'use them correctly'.
Strategy One: Use AI as a 'trend reference,' not a 'buy/sell signal'
AI predictions can help you understand the 'general direction,' for example:
- Is the overall property market in an upward cycle or an adjustment period?
- Which areas are experiencing an increase in supply?
- How do trends in mortgage interest rates affect repayment pressure?
But the specific timing of entering the market should still be determined based on your financial situation, property goals, and risk tolerance. AI is only a reference tool, not a basis for decision-making.
Strategy 2: Cross-Verify Multiple Data Sources
Don't just look at the prediction of one AI tool; you should compare:
- Market reports from traditional real estate agencies such as Centaline and Midland
- Official data from the Census and Statistics Department
- Market analysis from bank mortgage departments
- Valuation reports from professional surveying firms
The reliability of a prediction only increases when multiple data sources point in the same direction.
Strategy Three: Focus on 'Supply and Demand' Rather Than 'Price Forecasting'
Instead of asking 'Will housing prices go up or down next month,' it is better to ask:
- How many new developments will be completed in this area in the next three years?
- What is the current vacancy rate?
- Is the volume of second-hand listings increasing or decreasing?
- Are there any major infrastructure or facilities being completed in the area?
The supply and demand relationship is the core factor determining the long-term trend of property prices. AI can help you organize this data, but the analysis and judgment still rely on yourself.
Strategy 4: Combine Professional Opinions, Don't Go It Alone
Even if you use the most advanced AI tools, you should consult professional real estate agents, mortgage advisors, and surveyors. Their practical experience and market intuition cannot be replaced by AI.
I have seen too many clients make the wrong decisions at critical moments because of overconfidence (or over-reliance on AI). Remember: real estate investment is a team effort, not a solo game.
:::success Success case I have a client, Karen. She uses AI tools to identify areas with a 'supply shortage + supporting improvements,' and then asks me to conduct on-site inspections of specific residential complexes, followed by having a mortgage advisor calculate the optimal payment plan. This 'AI + professional team' combination allowed her to successfully enter the market at a low point in 2023 and see a 12% paper gain within six months. :::
Summary: AI is a tool, not a crystal ball
Returning to the question at the beginning of the article: Can AI really accurately predict next month's housing prices? The answer is: No.
AI can analyze historical data, find statistical patterns, and provide trend references, but it cannot predict sudden policy changes, cannot understand market sentiment, and cannot assess the quality of individual housing estates. More importantly, real estate investment is not about gambling, but about long-term planning.
If you are a first-time buyer preparing to get on the property ladder, instead of worrying about 'whether it will go up or down next month,' you might as well ask yourself:
- Is my financial situation stable?
- Does this unit meet my long-term needs?
- Is the contribution pressure within an affordable range?
- How are the amenities and development prospects of this area?
If you are an investor, you should pay more attention to 'supply and demand,' 'rental yield,' and 'capital appreciation potential,' rather than short-term price fluctuations.
Remember: Success in real estate investment has never depended on "accurate predictions," but on "correct strategies." AI can be your assistant, but the final decision-making power still rests in your hands.
Want to learn more about real estate investment strategies? Feel free to subscribe to our blog, where we share the latest market analysis and practical insights every week. If you have questions about a specific area or housing estate, you are also welcome to leave a comment below to discuss, or send us a private message to get professional advice.
Remember: In this long-term race of the real estate market, knowledge and experience are always more reliable than AI predictions. Let's work hard together to be smart investors!