AI in Real Estate Valuation: Transforming Property Appraisal
Explore the impact of AI on real estate valuation, distinguishing between Narrow AI and General AI in property appraisal. Learn how AI technology enhances the accuracy and efficiency of valuations through data-driven algorithms and models.
Download Presentation
Please find below an Image/Link to download the presentation.
The content on the website is provided AS IS for your information and personal use only. It may not be sold, licensed, or shared on other websites without obtaining consent from the author.If you encounter any issues during the download, it is possible that the publisher has removed the file from their server.
You are allowed to download the files provided on this website for personal or commercial use, subject to the condition that they are used lawfully. All files are the property of their respective owners.
The content on the website is provided AS IS for your information and personal use only. It may not be sold, licensed, or shared on other websites without obtaining consent from the author.
E N D
Presentation Transcript
AI in AVMs: how far we are from replacing AI in AVMs: how far we are from replacing Valuers Valuers? ? A Valuer s dialogue with Chat GPT by Konstantinos Pallis REV, REV-PME MoB TEGOVA
Thats a question that lies in the back of every Valuer s head nowadays! and is a bit provocative for a Valuer s conference ..
I will try to give an answer in the next 25-30mins. The topic is rather huge, and tons of literature is out there for the benefits of using AI So, I had to move in an effective, productive way, using trustworthy tools to search the international bibliography. In other words, I had to resort to AI. I did it using ChatGPT!
I started by asking the AI definition and here is the answer: and here is the answer: Artificial Intelligence (AI) refers to the simulation of human intelligence in machines that are designed to think and act like humans. AI systems can perform tasks that typically require human intelligence, such as understanding natural language, recognizing patterns, solving problems, learning from experience, and making decisions. AI can be categorized into two main types: Narrow AI: This type of AI is designed to perform a specific task or a limited range of tasks. Examples include voice assistants like Siri or Alexa, recommendation systems, and image recognition software. General AI: This is a theoretical form of AI that would possess the ability to understand, learn, and apply intelligence across a wide range of tasks, similar to a human being. As of now, general AI has not been achieved. AI technologies include machine learning, neural networks, natural language processing, computer vision, and robotics, among others.
Now a new question arises out of it: Where valuation belongs? Narrow or General AI? and here is the answer: and here is the answer: Valuation of real estate typically falls under the category of Narrow AI. This is because real estate valuation involves specific tasks such as analyzing market trends, assessing property features, and comparing similar properties, which can be effectively addressed using data-driven algorithms and models. Narrow AI is capable of performing these specialized functions by leveraging techniques such as machine learning and statistical analysis to provide property valuations based on various factors. However, it does not possess the general understanding and reasoning abilities that would be characteristic of General AI.
Valuation of real estate typically falls under the category of Narrow AI. This is because real estate valuation involves specific tasks such as analyzing market trends, assessing property features, and comparing similar properties, which can be effectively addressed using data-driven algorithms and models. So, we are back to the basics of valuations -Data collection, -Analysis of market trends, -Property features -Comparison of similar properties etc etc. However, the word typically used, leaves room for something different in some cases
But there is more to that: Narrow AI is capable of performing these specialized functions by leveraging techniques such as machine learning and statistical analysis to provide property valuations based on various factors. However, it does not possess the general understanding and reasoning abilities that would be characteristic of General AI. So can we live in valuation industry without the general understanding and reasoning abilities needed by a Valuer, in order to reach a meaningful decision about value in complex and nontransparent environments (markets) where there is usually lack of quality data?
So, I raised the question: how AI enhances real estate automated valuation models? and here is the answer: and here is the answer: AI enhances real estate automated valuation models by improving the accuracy and reliability of property valuations. AI algorithms can analyze vast amounts of data including property characteristics, local market trends, and comparable sales to generate more precise valuations. AI can also adapt and learn from past data to make better predictions and adjustments in real-time. Overall, AI helps to streamline the valuation process, reduce errors, and provide more informed and data-driven valuations for real estate properties. But what happens when we do not dispose even adequate amounts of quality data?
This answer gave me the hint for my next question to the system: how this can be done in opaque markets where sales prices and often properties characteristics are not known? and here is the answer: In opaque markets where sales prices are not readily available, AI can still enhance real estate automated valuation models by utilizing alternative data sources and advanced analytical techniques. Some ways AI can address this challenge include: 1. Using predictive modeling: AI algorithms can analyze various factors such as property characteristics, location, neighborhood trends, rental income, and demographic data to estimate property values in the absence of actual sales prices. (Comment: provided a strong correlation can be proved for these characteristics with assets values .) 2. Incorporating machine learning: AI can use historical data and real-time market information to train machine learning models that can make accurate predictions even in opaque markets. These models can continuously learn and improve over time as more data becomes available. 3. Leveraging satellite imagery and aerial data: AI can analyze satellite imagery and aerial data to assess property conditions, proximity to amenities, and other factors that can impact property values. This data can provide valuable insights in markets where sales prices are not transparent. (Comment: sounds like a whish list and not very pragmatic in real world .) 4. Utilizing alternative data sources: AI can gather and analyze alternative data sources such as rental rates, vacancy rates, local economic indicators, and social media sentiment to make more informed valuations in opaque markets. By combining these approaches, AI can enhance real estate automated valuation models in opaque markets and provide more accurate and reliable property valuations. Strange but no reference to asking prices is made here!
So, I moved to my next question: are asking prices treated with a discount to simulate transaction prices? and here is the answer: 1. In some cases, asking prices may be treated with a discount in order to simulate transaction prices in real estate automated valuation models. This approach is often used when actual sales prices are not available or when there is a significant time lag between when a property is listed and when it is sold. One common method is to apply a discount factor to asking prices based on historical data and market trends. This discount factor can vary depending on factors such as property type, location, time on the market, and other relevant variables. By adjusting asking prices with a discount, the model can better approximate the true market value of the property. However, it's important to note that this approach may not always be accurate, as asking prices may not always reflect the true market value of a property. Other factors such as negotiation tactics, seller motivation, and market conditions can also influence the final sales price. Therefore, while using asking prices with a discount can be a useful technique in the absence of transaction prices, it should be used cautiously and in conjunction with other data sources to ensure the accuracy of the valuation model. 2. Yes, in AI-assisted Automated Valuation Models (AVMs), asking prices can be treated with adjustments or discounts to better represent transaction prices. This is often done because the asking prices do not always reflect the actual market value or the price at which properties are ultimately sold. AVMs typically analyze various data sources, including recent sales prices, market trends, property characteristics, and location data. By applying statistical methods, they can adjust asking prices to estimate more accurate transaction prices. This helps to provide a more reliable and realistic valuation for real estate properties based on current market conditions.
Obviously, I was tempted to ask next question: how is the discount factor calculated? and here is the answer: The discount factor in Automated Valuation Models (AVMs) can be calculated using various methods, depending on the specific model and data available. Here are some common approaches: Historical Sales Data: By analyzing historical sales data, AVMs can determine the average difference between asking prices and final sale prices. This average difference can be used to establish a discount factor that is applied to current asking prices. Market Trends: AVMs may consider broader market trends, such as changes in supply and demand, economic conditions, and regional market performance. A declining market may require a larger discount factor compared to a stable or rising market. Comparative Market Analysis (CMA): AVMs often utilize a comparative approach where properties similar to the one being evaluated are analyzed. By looking at the ratios of final sale prices to asking prices for these comparable properties, a discount factor can be derived. Regression Analysis: Statistical regression techniques can help identify the relationship between various property features (such as size, location, age, and condition) and the price discrepancies between asking and transaction prices. This analysis can provide insights into appropriate discount rates to apply. (Comment: a lot of good quality data needed for this type of analysis ) Machine Learning Models: Advanced AVMs may employ machine learning algorithms that learn from large datasets of transactions to predict the most accurate discount factor based on the patterns observed in the data. The resulting discount factor is then applied to the asking price to estimate the transaction price. It s important to note that this factor can vary between different markets and property types.
Where are we so far? Where are we so far? AI is analyzing probably or surely in a more efficient way BIG amounts of data that are somehow made available to the AI software in order to identify market trends, assessing property features and their correlation with value, and comparing similar properties, which can be effectively addressed using data- driven algorithms and models. All good and meaningful so far BUT how are treated areas with few data available? here is the answer I received:
In areas with few data available, AVMs (Automated Valuation Models) use several strategies to provide valuations, as limited data can challenge the accuracy of estimates. Here are some common approaches: Use of Broader Geographic Regions: When local data is scarce, AVMs may extend their analysis to include comparable properties from nearby neighborhoods or regions that have sufficient sales data. This approach helps create a larger dataset for analysis. (Comment: this can lead to very inaccurate value assessment ) Statistical Techniques: Techniques such as spatial interpolation or statistical modeling can be applied. These methods help estimate property values based on the values of nearby properties, even if the nearby properties are not exactly the same. (Comment: this can lead to very inaccurate value assessment ) Market Segmentation: AVMs may segment the market into categories based on property types, sizes, and other characteristics. They can then apply a relevant market adjustment based on similar segments that might have more data available. (Comment: sounds like a wish list item .) Incorporating Non-Traditional Data: In the absence of sufficient sales data, AVMs may incorporate other relevant data points, such as listings, previous sales history, property tax assessments, and demographic data. This can provide additional context for estimating property values. Expert Override: In some cases, especially in highly variable markets, human expertise may be employed to adjust the model's outputs. Real estate professionals may review AVM results and tweak values based on their market knowledge and experience. (Comment: this means that we are entering General AI territory?)
In areas with few data available, AVMs (Automated Valuation Models) use several strategies to provide valuations, as limited data can challenge the accuracy of estimates. Here are some common approaches: Continued Application of Assumptions: AVMs may use standardized assumptions based on known factors about the area, such as average growth rates, typical price movements, or economic indicators. This approach provides a baseline that can be adjusted as more data becomes available. Confidence Intervals and Risk Assessment: AVMs may report a broader range of potential values or confidence intervals when data is limited, indicating the uncertainty in the valuation due to the lack of sufficient data. (Comment: in other words, less accurate valuations ) Using these strategies helps ensure that valuations in low-data areas remain as accurate as possible, although they may still carry higher uncertainty than valuations based on more robust datasets.
At this point I shall offer my personal experience from an AVM development project I completed in 2022 We developed in collaboration with the Statistics Dpt. of Athens University, an AVM for residential properties including four basic tools: Multiple regression Hedonic model Neuronic model Comparable approach (imitating the valuer s work but with some limitations) Regarding available data, we segmented the country in three parts, based on the availability of data: Athens 14.000 observations available Thessaloniki 3.500 observations available Rest of Greece 12.400 observations available
At this point I shall offer my personal experience from an AVM development project I completed in 2022 Continued We used approx. 30.000 data from valuations performed the three immediately passed years. The four tools used were giving various accuracy outputs and the value was calculated as a weighted average of the output of the four tools. All in all, we obtained the following accuracy levels:
At this point I shall offer my personal experience from an AVM development project I completed in 2022 regression hedonic neuronic comparative combined regression hedonic neuronic comparative combined regression hedonic neuronic comparative combined median 60%ile 70%ile 80%ile 90%ile 0.195 0.249 0.315 0.232 0.282 0.334 0.232 0.282 0.334 0.132 0.165 0.207 0.152 0.189 0.241 0.217 0.276 0.360 0.258 0.312 0.368 0.258 0.312 0.368 0.145 0.180 0.227 0.135 0.174 0.219 0.285 0.361 0.458 N/A N/A N/A 0.324 0.417 0.545 0.188 0.249 0.339 0.229 0.299 0.394 0.417 0.392 0.392 0.270 0.308 0.484 0.442 0.442 0.302 0.280 0.599 N/A 0.756 0.495 0.536 0.623 0.486 0.486 0.445 0.438 0.691 0.541 0.541 0.460 0.397 0.875 N/A 1.209 0.818 0.906 ATHENS THESSALONIKI REST OF GREECE
As a result, from the application of this AVM model we concluded the following: The comparative approach was giving the best accuracy where data were available In the specific AVM and data set used, comparative was used in 54,7% of the Athens area properties, 64,7% for Thessaloniki area properties and 69,5% for the rest of Greece properties. The lack of adequate granularity in the data used forced us to make assumptions that undermined the accuracy of the model. The scarcity of property data in the rest of Greece (lack of density) led the model to fail in this area, despite the sophisticated tools used.
As a result, from the application of this AVM model we concluded the following: continued There is potential for the model, provided that, we will be able to feed more data primarily for the rest of Greece and secondly for the big metropolitan areas of Athens and Thessaloniki. The accuracy of the model is considered useful for Athens and Thessaloniki but not useful for the rest of Greece. The lack of quality property data in general (typical for opaque markets, such as the Greek R.E. market) is undermining the results of the AVM, because is imposing the need for assumptions that are not always representatives for individual properties. Note: the model was developed for apartment type residential properties and not more complex assets (commercial etc.).
As a result, from the application of this AVM model we concluded the following: continued Quantity of data is of equal importance with granularity of data, especially regarding the characteristics of each property. Typical characteristics (independent variables)needed in our model were: Distance from transportation points Distance from attraction points Room number Property size EPC class AirConditioning Heating energy source Crime rate in the region Condition of maintenance Quality of construction Floor level Parking space Storage area Elevator Marketability Age of property etc etc. The need for assumptions re. these characteristics which are not always representatives for individual properties are undermining the accuracy of the model. This is critical too for the quality of the properties data included in a portfolio that is to be valued using an AVM.
So, I am coming back to my initial question: So, I am coming back to my initial question: AI in AVMs: how far we are from replacing Valuers? AI in AVMs: how far we are from replacing Valuers? From the analysis done so far, I think that we can safely conclude the following: There is no substitute for a large amount of good quality granular data. AI assisted AVMs can be useful in areas offering large amounts of good quality data as an extra aid to valuers. AI assisted AVMs are giving low accuracy value assessments in areas with a few and lower quality data typical of opaque markets to the extend that are not considered useful. The way towards implementation of AI assisted AVMs pass through the gathering of good quality data in large (statistically adequate) numbers, with sufficient granularity for the models to use them.
So, I am coming back to my initial question: So, I am coming back to my initial question: AI in AVMs: how far we are from replacing Valuers? AI in AVMs: how far we are from replacing Valuers? continued a qualitative demonstration can be the following a qualitative demonstration can be the following Amount / quality of data General AI non feasible Narrow AI feasible
So, the answer to my initial question: So, the answer to my initial question: AI in AVMs: how far we are from replacing Valuers? AI in AVMs: how far we are from replacing Valuers? I think that can be twofold : It will be very difficult or even impossible to replace valuers for an accurate assessment of value, because even if we have an accurate AI assisted AVM, its output will always need human review and verification, as it depends in various limitations / assumptions used by the model that can vary on a case-by-case basis. In opaque markets we are as far as we are from the creation of an open, transparent, well documented real estate market and the cultural / legal changes that are required for this for the creation of dependable AI assisted AVMs.
So, the answer to my initial question: So, the answer to my initial question: AI in AVMs: how far we are from replacing Valuers? AI in AVMs: how far we are from replacing Valuers? continued That is the reason why the EU issued the new CRR with the following description in Art. 229 regarding the valuation of a property: 1. The valuation of immovable property shall meet all of the following requirements: (a) the value is appraised independently from an institution s mortgage acquisition, loan processing and loan decision process by an independent valuer who possesses the necessary qualifications, ability and experience to execute a valuation; (b) the value is appraised using prudently conservative valuation criteria which meet all of the following requirements: (i) the value excludes expectations on price increases; (ii) the value is adjusted to take into account the potential for the current market value to be significantly above the value that would be sustainable over the life of the loan; (c) the value is documented in a transparent and clear manner; (d) the value is not higher than a market value for the immovable property where such market value can be determined; (e) where the property is revalued, the property value does not exceed the average value measured for that property, or for a comparable property over the last six years for residential property or eight years for commercial immovable property or the value at origination, whichever is higher
THANK YOU www.tegova.org
