Sixth Annual Pacific-Rim Real Estate Society Conference
`
`Sydney, Australia, 24-27 January 2000
`
`Using Expert Systems and Artificial Intelligence
`For Real Estate Forecasting
`
`Peter Rossini
`
`Lecturer
`
`School of International Business, University of South Australia
`
`Centre for Land Economics and Real Estate Research (CLEARER)
`
`Phone: 61-8-8302-0649, Facsimile: 61-8-83020512, E-mail: peter.rossini@unisa.edu.au
`
`Keywords: Real Estate Markets, Forecasting, Time Series Analysis, Neural Networks, Artificial
`Intelligence
`
`Abstract: This paper examines the use of expert systems and artificial intelligence, (in particular the
`application of neural networks) to real estate forecasting. While there is a great deal of literature about the use of
`artificial intelligence for mass appraisal, there is relatively little work on how it can be applied in real estate
`forecasting. This paper examines the current uses of artificial intelligence, particularly neural networks, in the
`business-forecasting field and considers suitable applications in real estate. The paper also considers the broader
`issue of expert systems and how a better system can lead to better results. Some real estate data are used as
`simple case studies to demonstrate their use.
`Introduction:
`
`Over the last few decades there have been significant changes to the methods of forecasting available to analysts
`and practitioners. Complex methods have become available for routine use and complex econometric models
`are often suggested as the solution to forecasting problems. However some researchers suggest that the use of
`better systems rather than better forecasting techniques would lead to better overall forecasts. This idea was
`strongly supported by the work of Makridakis et al. (1982). This research involved forecasting 1001 different
`time series using 24 different methods. They concluded that more sophisticated methods may produce no better
`results from simple ones. These are highlighted in the following quote from their conclusions.
`
`“If the forecasting user can discriminate in his choice of methods depending upon the type of data (yearly, quarterly,
`monthly), the type of series (macro, micro, etc.) and the time horizon of forecasting, then he or she could do
`considerably better than using a single method across all situations - assuming, of course, that the results of the
`present study can be generalized.... Even though further research will be necessary to provide us with more specific
`reasons as to why this is happening, a hypothesis may be advanced at this point stating that statistically sophisticated
`methods do not do better than simple methods (such as deseasonalized exponential smoothing) when there is
`considerable randomness in the data.... Finally, it seems that seasonal patterns can be predicted equally well by both
`simple and statistically sophisticated methods.”
`
`One implication of this is that forecasting systems with simple artificial intelligence (AI) and expert systems
`(ES) may produce better outcomes and be more efficient than those use a single, (and often more complex and
`sophisticated) model (DeLurgio, 1998). This paper examines some of the uses of expert systems and artificial
`intelligence within business and how they are being applied to real estate problems.
`
`1
`
`TRULIA - EXHIBIT 1011
`
`

`
`One particular aspect that is considered is the advantages of these systems as a learning tool for inexperienced
`real estate practitioners. An example is used to show how an expert system for residential valuations might work.
`
`What are expert systems and artificial intelligence?
`The field of artificial intelligence (AI) has developed rapidly as computing power has increased. Artificial
`intelligence refers to the ability to perform the intelligent functions of the human brain. In particular some forms
`of reasoning, some learning and general improvement over time. The uses of AI are varied with the major uses
`so far being in the computing and robotics area. They form an integral part in modern optical character and
`speech recognition software, are broadly used in robotics and have very wide spread applications through the
`military. The use of AI is now extending into the social sciences including business studies. The use of artificial
`neural networks (ANN) and genetic algorithms are becoming more wide spread particularly in the fields of
`market research and forecasting.
`
`Expert systems may be considered to be a subset of AI. DeLurgio (1998) makes a clear distinction between
`conventional program systems (CPS) and expert systems (ES). He maintains that CPS involves the researcher
`wanting to create a system that deals with interesting and difficult tasks without regard to whether these are
`similar to those used by humans i.e. it does not matter how the job gets done, as long as it does. The ES tries to
`gain an understanding of how humans solve problems and then uses the computer to explain and predict their
`behavior. In practice many systems contain elements of both. So that many systems have some aspects of
`expert systems but often rely on some of the basic number crunching abilities of a CPS. This will increasingly
`become the trend in real estate applications where hybrid techniques will become more prominent (McCluskey,
`1999). The emergence of expert or partly expert systems is important for educators in nearly all fields. The
`advantages of ES include the ability to provide expert advice to non-experts, assist experts to solve problems and
`to act as a teaching tool for non-experts (DeLurgio, 1998). For educators the final issue is very beneficial. A
`well-constructed ES can form a valuable teaching and training tool.
`
`The use of artificial intelligence for forecasting
`The most used AI technique is probably artificial neural networks (ANN). The concept of the ANN is that of a
`learning algorithm similar to the function of the human brain. They work by a series of interconnected neurons
`in a similar manner to the working of the brain. However even with the largest modern computers it is estimated
`that an ANN with 10 million interconnections would have a neuron structure somewhat smaller than a
`cockroach. (De Lurgio, 1998). The process of using the ANN for forecasting is largely the same as for other
`forecasting methods such as multiple regression. As a results these two techniques are very often compared. In
`each case there is input data which is used to model output data. They each use a series of coefficients in the
`modeling process and each attempt to minimize error terms in a similar manner. The standard methods of hold
`out samples are also commonly used in both as a measure of the forecasting ability. The internal process of the
`ANN is however more complex and less easy to reproduce and explain. It functions as a “black box” to a much
`larger extent than for traditional statistical methods. On the other hand, people with no background in the
`method seem to be able to make better predictions using ANN’s. This sets a dangerous precedent and it is
`probable the use of ANN’s will be over-sold and they will be used in situations where more conventional
`methods are probably superior. As a result, dangerous conclusions and recommendations will be made by
`people who use ANN’s badly. Notwithstanding this ANN’s have been well researched in business fields in
`recent years. For a basic time series situation Kuo et al. (1996) found that neural networks produced lower errors
`than Box-Jenkins and regression procedures. Denton (1995) found ANN to be superior for causal forecasting to
`regression. There are numerous examples of where ANN’s have been used for business forecasting. These
`include forecasting of electricity consumption (Nizami et al., 1995), airline passengers (Man et al 1995),
`company audits (Lanard et al (1995), bank failures (Tam, 1992), bankruptcies (Fletcher, 1993), stocks and bonds
`(Desai, 1998, Li, 1994), futures and financial markets (Meade, 1995, Kaastra et al, 1995, Mangasarian, 1995,
`Kuan et al, 1995, Grudnitski et al, 1993). In most cases researchers have found that ANN’s can produce
`forecasts with lower overall errors than with conventional methods such as regression.
`
`2
`
`

`
`The use of artificial intelligence for real estate forecasting
`
`Forecasting is a major issue in most aspects of real estate practice. Valuation and appraisal are forecasting.
`Property development relies on forecasting of expected costs and returns. Property and facilities managers use
`forecasts of supply and demand as well as of costs and returns. Funds and investment managers rely on forecasts
`of value now and in the future through forecasts of growth and economic activity. With all this forecasting being
`relied upon it is somewhat surprising that the uses of AI and ES are primarily restricted to mass appraisal,
`however this is less surprising when an analysis of the use suggests that most of this would fit better into the
`description of conventional program systems. Early attempts at “automating” or “computer assisting” valuation
`go back as far as the late 1970’s when sufficient computing power became available (Eckert, 1993, Jensen,
`1984). The use of expert systems and artificial intelligence techniques for residential valuation has been
`suggested in the literature for over a decade. Methods such as rule-based reasoning (Scott et al. 1989, Nawawi et
`al. 1997), case-based reasoning (O’Roarty et al. 1997), and neural network (Borst 1995, Do et al. 1992, Evans et
`al. 1993, James, 1996, Jensen 1990, Lenk et al., McClusky et al. 1996, Rayburn, 1995, Rossini 1997, Tay and
`Ho 1994, Worzala 1995) have all been suggested as means of approaching mass appraisal and to some extent
`valuation generally. In general the emphasis has been on data mining from a large property transaction database.
`There are a wide variety of methods that can be used for data mining but these can be classified into nine groups;
`classification, regression, discovery of associations, discovery of sequential patterns, temporal modeling,
`deviation detection, dependency modeling, clustering and characteristic rule discovery (McCluskey and Anand,
`1999).
`
`The use of AI through neural networks is not well researched in other aspects of property. Kershaw et al (1999)
`compared the results of neural networks to those using regression to develop time series indices for residential
`data. Neural networks were found to be useful (but no better) for estimating a hedonic price index based on
`cross sectional transaction data but were found to be quite useful when dealing with time series data e.g. median
`prices. Other examples of the use of neural networks have been in the property development/ building fields
`including models for the demand for residential construction (Hua, 1996) and cost estimation. (De la
`Garza,1995).
`
`Using of artificial intelligence and expert systems in real estate practice
`The discussion so far has focused around neural networks and data mining techniques for forecasting.
`Forecasting is primarily a quantitative process using numerical data from the past to forecast the future. Expert
`systems are an ideal method for dealing with many other problems as well. One of these is qualitative
`forecasting. This is typically used for new products and in situations where there is no long-term series that
`might assist in giving a forecast. Expert systems can assist in the use of methods such as sales-force composites,
`surveys of customers and populations, jury of executive opinion and the Delphi method (Wilson et al, 1998).
`There are numerous other areas within real estate practice where expert systems could be usefully employed.
`Some examples are
`•
`
`the preparation of real estate documentation such as leases, contracts and forms. Expert systems can help
`the user prepare better documents by guiding them through the process and brining attention to issues that
`might other wise have been missed, as well as improving phrasing and structure.
`
`•
`
`•
`•
`•
`
`the costing of buildings and development projects. This requires a combination of data mining and rule
`based methods (or case based reasoning) to bring together the expert knowledge of quantity surveying,
`engineering and construction with current cost estimates.
`
`computing assistance with specialised as well as general computer software.
`
`preparation of reports and property descriptions
`
`property and facilities management problems where the ES can be used by both clients and property
`managers to streamline the solution to some problems
`
`One of the obvious issues with each of these is that they are useful for the novice and can therefore assist in the
`teaching and training process until the user becomes and expert themselves.
`
`3
`
`

`
`Example of a common rule based expert system
`
`Many useful expert systems use a rule-based approach where a set of rules are established and the user
`effectively moves from a starting point to some answer or output by answering a set of questions. Each next
`question is dependent upon the last answer or series of answers. There are many thousands of examples of such
`systems, and they are now widely used. shows a simple rule based system from the British Medial Association
`Family Health Encyclopedia (1997). Listed as interactive diagnostic charts, the program used a flow chart
`metaphor to assist the user to self-diagnose problems. The example in shows only the first of a large number of
`rule-based questions that eventually lead the user to a simple home remedy or advice to seek further expert help.
`More complex versions of this type of program are routinely used by the medical profession for education and
`training and as a continuous and updated reference.
`
`Figure 1 - Example of a simple expert system (British Medial Association Family Health Encyclopedia)
`
`An example of a residential valuation system
`
`Valuation systems that are being proposed are usually more in the CPS rather than the ES mould. Most are
`probably better termed as automated or computer assisted valuation tools. The simplest use purely rule-based
`systems , sometimes implementing cost based methodologies. Future expert systems that utilise artificial
`intelligence, are likely to use hybrid techniques (McCluskey, 1999). The example discussed here had its
`foundation in 1992 (Rossini et al., 1992) with the design of a basic computer assisted appraisal system using
`Microsoft Windows. Since then it has progressed but is still clearly a “half way system” between CPS and ES,
`with future developments being in the ES area. The system uses a series of steps that the developers believe to
`automate a sound residential valuation practice with expert inputs and advice but with the option to override the
`system at any stage. Preliminary research using the (somewhat awkward) prototype system suggest that the
`accuracy of such a system would more than match the industry standards for normal individual residential
`valuations. (Rossini, 1999).
`
`4
`
`

`
`The system uses a seven-step process
`
`Step 1.
`
`Step 2.
`
`Collect and input details of the subject property.
`
`Find “appropriate” sales from the market place - using search and filter systems from the monthly
`updated sales database.
`
`Step 3. Model the market using sales from Step 2. Test for any relationships and find the coefficients.
`
`Step 4.
`
`Step 5.
`
`Establish major value determinants and adjustments from the model in Step 3.
`
`Find “appropriate” number of the most comparable sales - using expert knowledge and value
`determinants from Step 4.
`
`Step 6. Make adjustments to the most comparable sales to allow for differences between the subject and sale
`properties.
`
`Estimate value based on adjusted prices and relative comparability of each sale - using a weighted
`mean approach.
`
`Step 7.
`
`Step 1
`
`The first step is to collect data about the subject property. For advanced or experienced users this data can be
`input directly. For the novice user or trainees, there is a rule-based system to allow for the user to be guided to
`the inputs. The prototype-input screen is shown as Figure 2. In the basic prototype this screen is used to input
`data, select methods and output the answer. The system will work with or without the data listed in the
`inspection report. This enable a valuation based on the “features” which are listed on the state valuation list
`however the result is less accurate in many locations.
`
`Figure 2 - Step 1 of the Valuation System - inputting the data
`
`5
`
`

`
`Step 2
`
`The second step involves collecting sales data from the sales database. The data is updated monthly and the
`system utilises an existing search and filter system together with expert knowledge of the appropriate parameters
`to select the most sales set. These parameters can be over ridden by the user and the next version of the system
`will remember what parameters were effective in each location.
`
`Figure 3 - Step 2 of the Valuation System - collecting sales data
`
`Step 3
`
`The third step is to analyse the market data. In this example an additive regression model is used. The system
`allows for a variety of different models that include logged regression models and various neural network
`models with genetic algorithms. The model building process is based on expert knowledge with standard tests
`being used for model validity. In Figure 3 only the land area, building area, building age and a style dummy
`variable are used. With different data, different appropriate models are found depending upon the location and
`the physical and economic environment.
`
`Figure 4 - Step 3 of the Valuation System - modeling the data
`
`S U M M A R Y O U T P U T
`
`Regression Statistics
`Multiple R
`0.986244
`R Square
`0.972677
`Adjusted R Square
`0.968774
`Standard Error
`5527.687
`Observations
`33
`
`A N O V A
`
`Regression
`Residual
`Total
`
`Intercept
`Land Area
`Equiv Area
`Year built
`Conventional
`
`df
`
`M S
`SS
`4 3.05E+10 7.61E+09
`28 8.56E+08 30555325
`32 3.13E+10
`
`Significance F
`F
`249.198
`1.89E-21
`
`CoefficientsStandard Error
`-2627631
`288667.5
`690987.6
`52798.42
`599.8755
`32.88361
`1327.624
`146.5606
`-4750.27
`2475.67
`
`t Stat
`-9.10262
`13.08728
`18.24239
`9.058533
`-1.91878
`
`P-value
`7.35E-10
`1.87E-13
`4.46E-17
`8.15E-10
`0.065259
`
`Lower 95% Upper 95%
`-3218940 -2036322
`582834.8
`799140.4
`532.5164
`667.2346
`1027.408
`1627.841
`-9821.46
`320.9122
`
`6
`
`

`
`Step 4
`
`Figure 5 - Step 4 of the Valuation System - quantifying the adjustments
`
`Major Value Determinants
`Land Area
`Equivalent Building Area
`Year of Construction
`Building Style
`
`Value Adjustments
`Land Area
`69$
`
`Equiv Area
`
`600$
`Year built
`
`1,328$
`Conventional
`
`4,750-$
`
`Per Sq Metre
`Per Sq Metre
`Per Year
`If Conventional
`
`The model of the market data is used as the
`basis for the adjustments for the system.
`This example using an additive regression
`model uses an additive adjustment method
`although other methods are available. The
`user can override the adjustments. In Figure
`5 there is no adjustment for date of sale
`because the model of the market did not find
`any measurable impact of time on prices. If
`the user believes that there has been a very
`recent increase in prices then a suitable
`adjustment can be made.
`
`Step 5
`The next stage involves selecting the most comparable sales. This process uses a combination of expert
`knowledge and nearest neighbour techniques. The system attempts to find an appropriate number of comparable
`sales. This involves establishing the “jump” in a comparability rating eg if there are three quite comparable sales
`but the fourth sale is much less comparable, then only three will be selected. As with all parts of the system, the
`user can override the automatic system and select the most comparable sales.
`
`Figure 6 - Step 5 of the Valuation System – select the most comparable sales
`
`Address
`SUBJECT
`17 KARRI DR
`13 ACACIA AVE 11 CHERINGAR BLVD9 ORANGE GROVE CCT1A TURNER TCE 47 BALMORAL RD
`$113,000
`$106,500
`$116,500
`$127,000
`$108,000
`$114,000
`?????
`Sale Price
`637
`796
`637
`719
`581
`729
`728
`Land Area
`150
`136
`136
`136
`145
`148
`143
`Equiv Area
`1963
`1960
`1973
`1984
`1971
`1965
`1962
`Year built
`12-05-97
`11-07-97
`26-06-97
`23-07-97
`28-02-97
`03-07-97
`Now
`Sale Date
`R1
`R1
`R1
`R1
`R1
`R1
`R1
`Zone
`BRICK
`BRICK
`BRICK
`BRICK
`BRICK
`BRICK
`BRICK
`Wall
`TILED
`TILED
`TILED
`TILED
`TILED
`TILED
`TILED
`Roof
`CONVENL CONVENL CONVENL CONVENL CONVENL CONVENL CONVENL
`Style
`6
`5
`5
`6
`6
`5
`6
`Rooms
`8
`6
`7
`8
`7
`7
`7
`Condition
`Improvements 6H DCP V 5H B/GAR CP 5H CP
`6H CP
`6H CP 5H CP SH R/R 6H CP V SP
`
`Step 6
`
`Step 6 utilises all the information collected to this point, to adjust the most comparable sales using the
`appropriate method and the selected adjustment factors. Figure 7 shows how these adjustments are made using
`the additive adjustment method used in this example.
`
`7
`
`

`
`Figure 7 - Step 6 of the Valuation System - applying the adjustments
`
`Address
`Sale Price
`Land Area
`Equiv Area
`Year built
`
`SUBJECT
`?????
`728
`143
`1962
`
`13 ACACIA AVE 11 CHERINGAR BLVD9 ORANGE GROVE CCT1A TURNER TCE 47 BALMORAL RD
`17 KARRI DR
`$106,500 $116,500 $127,000 $108,000 $114,000 $113,000
`796
`637
`719
`581
`729
`637
`136
`136
`136
`145
`148
`150
`1960
`1973
`1984
`1971
`1965
`1963
`
`Adjustments
`Sale Price
`Land Area
`Equiv Area
`Year built
`Adjusted Sale Price
`
`$106,500
`$116,500
`$127,000
`$108,000
`$114,000
`$113,000
`-4692
`6279
`621
`10143
`-69
`6279
`4200
`4200
`4200
`-1200
`-3000
`-4200
`2656
`-14608
`-29216
`-11952
`-3984
`-1328
`$108,664 $112,371 $102,605 $104,991 $106,947 $113,751
`
`Step 7
`The final step involves weighting the adjusted sales prices on the basis of comparability and calculating a
`weighted average. A mathematical “nearness” calculation is uses to estimate these weights but expert
`knowledge and a rule-based system will assist the user to override these with what may well be more appropriate
`subjective value. This can be particularly useful when trying to allow for some aspects of comparability that are
`difficult to quantify. For example if the subject property has an outstanding view, then sales with a similar view
`may be considered most comparable since other more quantifiable factors are likely to be adjusted for.
`
`Figure 8 - Step 7 of the Valuation System - weighting the adjusted sales and estimating a value
`
`Adjusted Sale Price
`Comparability Weight
`
`$108,656
`20.4%
`
`$112,383
`9.4%
`
`$102,613
`6.9%
`
`$105,009
`10.1%
`
`$106,949
`33.4%
`
`$113,761
`19.9%
`
`Weighted Mean Value
`
`$108,666
`
`Throughout the process the system will offer advice about the appropriateness of the outcomes. In some
`situations the system will not ever provide suitable outcomes and the user will be advised of this.
`
`This system is designed overall to take the user through the valuation process. Users should become better at
`valuation through the process. One of the major advantages of the system, and a probable reason for its inherent
`accuracy compared to manual valuations, is the requirement for the user to follow all the steps. Using this
`system the user must consciously decide to take a shortcut such as ignoring recent sales and what this implies
`about the market place.
`
`Conclusion
`
`This paper has examined some of the current issues regarding the use of expert systems and artificial intelligence
`for practitioners in the real estate industry. The use of AI and ES have been suggested for a wider range of uses
`than is currently being research to any great extent. An important concept of the use of ES is that it is the overall
`system that is important rather than the reliance on new and “better” techniques. A valuation system that is
`approaching an expert system as examined as an example of how a system can work and how it can be used as a
`useful tool in teaching and training.
`
`8
`
`

`
`References
`Borst, R. (1995) “Artificial neural networks in mass appraisal”, Journal of Property Tax Assessment &
`Administration, 1(2): 5-15
`
`British Medical Association Family Health Encyclopaedia [CD Rom], (1997) Doring Kindersley, London
`
`De la Garza J and Rouhana K. (1995) "Neural Networks versus Parameter-Based Applications in Cost
`Estimating." Cost Engineering 37, no. 2 (February 1995), pp. 14-18.
`
`DeLurgio (1998) Forecasting Principles and Applications, McGraw-Hill International, USA
`
`Denton J (1995) “How Good Are Neural Networks For Causal Forecasting?” Journal of Business Forecasting
`(Summer 1995), pp. 17-20.
`
`Desai V. and Bharati R. (1998) “The Efficacy of Neural Networks in Predicting Returns on Stock and Bond
`Indices”, Decision Sciences Vol. 29. No.2 Spring 1998, pp. 405-425.
`
`Do, A. and Grudnitski, G. (1992), “A Neural Network Approach to Residential Property Appraisal”, The Real
`Estate Appraiser, Dec. 1992:38-45
`
`Eckert J., O’Connor P. and Chamberlain C (1993) “Computer-Assisted Real Estate Appraisal: A California
`Savings and Loan Case Study” The Appraisal Journal, October 1993, 524-532
`
`Evans, A. James, H. And Collins, A. (1993), “Artificial Neural Networks: an Application to Residential
`Valuation in the UK”, Journal of Property Valuation & Investment: 11:195-204
`
`Fletcher D. and Goss E. (1993) “Forecasting with neural networks - An application using bankruptcy data”
`Information & Management Vol. 24 (1993) pp 159-167
`
`Grudnitski G. and Osburn L. (1993) “Forecasting S&P and Gold Futures Prices: An Application of Neural
`Networks” The Journal of Futures Markets. Vol. 13, No 6, pp 631-643
`
`Hua C. (1996) "Residential Construction Demand Forecasting Using Economic Indicators: A Comparative Study
`of Artificial Neural Networks and Multiple Regression." Construction Management and Economics 14.
`no. 1 (January 1996), pp. 125-34.
`
`James, H. And Lam, E, (1996) “The Reliability of Artificial Neural Networks for Property Data Analysis”, Third
`European Real Estate Society Conference, Belfast, June 26-28
`
`Jensen, D. (1984) “Alternative Modeling Techniques in Computer-Assisted Mass Appraisal”, Appraisal Journal
`
`Jensen, D. (1990) “Artificial Intelligence in Computer-Assisted Mass Appraisal”, Property Tax Journal, Vol. 9,
`5-26
`
`Kaastra I. and Boyd M. (1995) "Forecasting Futures Trading Volume Using Neural Networks." Journal of
`Futures Markets, 15, no. 8 (December 1995), pp. 953-70.
`
`Kershaw P. & Rossini, P. (1999) "Using Neural Networks to Estimate Constant Quality House Price Indices",
`International Real Estate Society Conference, Kuala Lumpur, 26-30 January, 1999
`
`Kuan C and Liu T. (1995) "Forecasting Exchange Rates Using Feedforward and Recurrent Neural Networks."
`Journal of Applied Econometrics 10, no. 4 (October-December 1995), pp. 347-64.
`
`Kuo C and Reitsch A. (1996) “Neural Networks v.s Conventional Methods of Forecasting”. Journal of Business
`Forecasting (Winter 1995/1996), pp. 17-22.
`
`Lenard M.; Alam P.; and Madey G.(1995) "The Application of Neural Networks and a Qualitative Response
`Model to the Auditor's Going Concern Uncertainty Decision." Decision Sciences 26, no. 2 (March-April
`1995), pp. 209-27.
`
`9
`
`

`
`Lenk, M., Worzala, E., and Silva, (1997) “High-tech valuation: should artificial neural networks bypass the
`human valuer?”, Journal of Property Valuation and Investment, Vol 15, No1, 1997, 8-26
`
`Li, E.Y. (1994) “Artificial neural networks and their business applications” Information & Management. Vol. 27
`(1994) pp 303-313
`
`Makridakis, S., Wheelwright S. and McGee V. (1982) “The Accuracy of Extrapolation (Time Series) Methods:
`Results of a Forecasting Competition” Journal of Forecasting, 1 – 1982, pp111-153.
`
`Mangasarian O. (1995) "Neural Networks in Finance and Investing". Interfaces 25, no. 1 (January-February
`1995), pp. 141-42.
`
`McCluskey W. and Anand S. (1999) “The application of intelligent hybrid techniques of residential properties”
`Journal of Property Investment & Finance. Vol. 17. No. 3, 1999 pp218-238
`
`McCluskey, W., Dyson, K., McFall, D. & Anand, S. (1996) “Mass Appraisal for Property Taxation: An
`Artificial Intelligence Approach”, Land Economics Review, Vol. 2, No 1, 25-32
`
`Meade N. (1995) "Neural Network Time Series Forecasting of Financial Markets." International Journal of
`Forecasting 11, no. 4 (December 1995), pp. 601-602.
`
`Nam K and Schaefer T. (1995) "Forecasting International Airline Passenger Traffic Using Neural Networks."
`Logistics and Transportation Review 3 1, no. 3 (September 1995), pp. 239-5 1.
`
`Nawawi, A.H., Jenkins, D. and Gronow, S. (1997), “Expert system development for the mass appraisal of
`commercial property in Malaysia” Journal of the Society of Surveying Technicians, Vol18 No.8, p. 66-72
`
`Nizami S and AI-Garni A. (1995)"Forecasting Electric Energy Consumption Using Neural Networks." Energy
`Policy 23, no. 12 (December 1995), pp. 1097-1104.
`
`O’Roarty, B., Patterson, D., McGreal, W.S., Adair, A.S. (1997) “A case based reasoning approach to the
`selection of comparable evidence for retail rent determination”, Expert Systems with Applications, Vol.
`12 No 4, pp. 417-28.
`
`Rayburn W. (1995) “Artificial Intelligence: The Future of Appraising”, The Appraisal Journal, October 1995,
`pp 429 -435
`
`Rossini P. (1997) “Artificial Neural Networks versus Multiple Regression in the Valuation of Residential
`Property”, Australian Land Economics Review, November 1997 Vol. 3 No 1
`
`Rossini P., Kershaw P. & Kooymans R. (1992) "MicroComputer Based Real Estate Decision-Making and
`Information Management - An Integrated Approach" Second Australasian Real Estate Educators
`Conference, Adelaide, 1992.
`Rossini, P. (1998) “Improving the Results of Artificial Neural Network Models for Residential Valuation”, 4th
`Pacific Rim Real Estate Society Conference, Perth 1998
`
`Rossini, P. (1999) “Accuracy Issues for Automated and Artificial Intelligent Residential Valuation Systems”,
`International Real Estate Society Conference, Kuala Lumpur, 26-30 January, 1999
`
`Scott, I. and Gronow, S. (1989), “Valuation expertise: its nature and application” Journal of Valuation, Vol 8,
`No4, p 362-375
`
`Tam K and Melody Y (1992) “Managerial Applications of Neural Networks: The Case of Bank Failure
`Predictions” Management Science, Vol. 38. No. 7, July 1992, pp 926-947.
`
`Tay, D. and Ho, D. (1994), “Intelligent Mass Appraisal”, Journal of Property Tax Assessment & Administration,
`Vol. 1, No 1, 5-25
`Wilson J. and Keating B. (1998) Business Forecasting, Irwin McGraw-Hill, USA, 3rd Edition
`
`Worzala, E., Lenk, M. and Silva, (1995) “An Exploration of Neural Networks and Its Application to Real Estate
`Valuation”, The Journal of Real Estate Research, Vol. 10 No. 2
`
`10