Real Estate Modelling and Forecasting Hardcover_15 docx

Real estate forecasting in practice 427an expert makes an adjustment to the forecast driven by future ment growth, this adjustment is based on a less efficient use of the his-torical rela

Real estate forecasting in practice 427

an expert makes an adjustment to the forecast driven by future ment growth, this adjustment is based on a less efficient use of the his-torical relationship between rent and employment growth The expertshould direct his/her efforts towards influences that will genuinely add

employ-to the forecast When the forecasts from a model and expert opinionbring different kinds of information and when the forecasts are not cor-related, it is beneficial to combine them (Sanders and Ritzman, 2001)

(2) Track record assessment Purely judgemental forecasts or adjusted model

forecasts should be evaluated in a similar manner to forecasts fromeconometric models The literature on this subject strongly suggeststhat track record is important It is the only way to show whether expertopinion is really beneficial and whether judgement leads to persistentoutperformance It provides trust in the capabilities of the expert andhelps the integration and mutual appreciation of knowledge betweenthe quantitative team and market experts Clements and Hendry (1998)assert that the secret to the successful use of econometric and timeseries models is to learn from past errors The same approach should

be followed for expert opinions By documenting the reasons for theforecasts, Goodwin (2000a) argues that this makes experts learn fromtheir past mistakes and control their level of unwarranted intervention

in the future It enables the expert to learn why some adjustmentsimprove forecasts while others do not As Franses (2006) notes, the bestway to do this is to assess the forecasts based on a track record

Do the experts look at how accurate their forecasts are, though? Fildesand Goodwin (2007) find that experts are apparently not too botheredabout whether their adjustments actually improve the forecasts Thisdoes not help credibility, and hence it is important to keep track records

(3) Transparency The way that the forecast is adjusted and the judgement

is produced must be transparent If it is unknown how the expert hasmodified the model, the forecast process is unclear and subjective

13.6 Integration of econometric and judgemental forecasts

The discussion in section 13.2 has made clear that there are benefits frombringing judgement into the forecast process As Makridakis, Wheelwrightand Hyndman (1998, p 503) put it: ‘The big challenge in arriving at accu-rate forecasts is to utilize the best aspects of statistical predictions whileexploiting the value of knowledge and judgmental information, while alsocapitalizing on the experience of top and other managers.’ The potentialbenefits of combining the forecasts are acknowledged by forecasters, and

428 Real Estate Modelling and Forecasting

this leads to the subject of how best to integrate model-based and tal forecasts The integration of econometric and judgemental forecasts is

judgemen-a well-resejudgemen-arched topic in business economics judgemen-and finjudgemen-ance In summjudgemen-ary,this literature points to different approaches to integrating econometricforecasts and judgemental views A useful account of how the forecasts arecombined is given by Timmermann (2006)

(1) Mechanical adjustments to the statistical forecast The forecast team may

inves-tigate whether gains can be made by mechanical adjustments to themodel’s forecasts in the light of recent errors For example, one suchprocedure is to take part of the error in forecasting the latest period(usually a half of the error) and add that to the forecast for the nextperiod Consider that a model of retail rents based on consumer spend-ing has over-predicted rent growth in the last few periods (fitted aboveactual values) This could be due to intense competition between retail-ers, affecting their turnover, that is not captured by the model Wemechanically adjust the first forecast point by deducting half the error

of the previous period or the average of the previous two periods andperhaps a quarter of the error of the following period (so that we lowerthe predicted rental growth) A prerequisite for this mechanical adjust-ment is, of course, our belief that the source of the error in the last fewobservations will remain in the forecast period Vere and Griffith (1995)have found supportive evidence for this method but McNees (1986) haschallenged it

(2) Combining judgemental and statistical forecasts produced independently Aside

from mechanical adjustment, another approach is to combine experts’judgemental forecasts with the estimates of a statistical method pro-duced separately It is assumed that these forecasts are produced inde-pendently; if the parties are aware of each other’s views, they mightanchor their forecasts This approach appears to work best when theerrors of these forecasts take opposite signs or they are negatively cor-related (note that a historical record may not be available), although it

is not unlikely that a consensus will be observed in the direction of thetwo sets of forecasts

A way to combine these forecasts is to take a straightforward average

of the judgemental and econometric forecasts (see Armstrong, 2001).More sophisticated methods can be used If a record of judgementalforecasts is kept then the combination can be produced on the basis ofpast accuracy; for example, a higher weight is attached to the methodthat recently led to more accurate forecasts As Goodwin (2005) remarks,

a large amount of data is required to perform this exercise, which thereal estate market definitely lacks

Real estate forecasting in practice 429

Goodwin also puts forward Theil’s correction to control judgementalforecasts for bias This also requires a long series of forecast evaluationdata Theil’s proposal is to take an expert’s forecasts and the actual valuesand fit a regression line to these data Such a regression may be

yield = 2 + 0.7 × judgemental yield forecast

In this regression, yield is the actual yield series over a sufficiently

long period of time to run a regression Assume that the target variable

yield refers to the yield at the end of the year Judgemental yield forecast

is the forecast that was made at, say, the beginning of each year Whenmaking the out-of-sample forecast, we can utilise the above regression

If the expert predicts a yield of 6 per cent, then the forecast yield is2%+ 0.7 × 6% = 6.2%

Goodwin (2000b) has found evidence suggesting that Theil’s methodworks It requires a long record of data to carry out this analysis, however,and, as such, its application to real estate is restricted Goodwin (2005)also raises the issue of who should combine the forecasts He suggeststhat the process is more effective if the user combines the forecasts Forexample, if the expert combines the forecasts and he/she is aware ofthe econometric forecasts, then the statistical forecast can be used as ananchor Of course, the expert might also be the user For further reading

on this subject, Franses (2006) proposes a tool to formalise the so-called

‘conjunct’ forecasts – that is, forecasts resulting from an adjustment bythe expert once he/she has seen the forecast

(3) The ‘house view’ This is a widely used forum to mediate forecasts and

agree the organisation’s final forecasts The statistical forecasts and thejudgemental input are combined, but this integration is not mechanical

or rule-based In the so-called ‘house view’ meetings to decide on thefinal forecasts, forecasters and experts sit together, bringing their views

to the table There is not really a formula as to how the final output will

be reached Again, in these meetings, intervention can be made based

on the practices we described earlier, including added factors, but theprocess is more interactive

Makridakis, Wheelwright and Hyndman (1998) provide an example

of a house view meeting The following description of the process drawsupon this study but is adapted to the real estate case The house viewprocess can be broken down into three steps

Step 1

The first step involves the preparation of the statistical (model-based)forecast This forecast is then presented to those attending the houseview meeting, who can represent different business units and seniority

430 Real Estate Modelling and Forecasting

Participants are given the statistical forecasts for, say, yields (in a ular market or across markets) This should be accompanied by an expla-nation of what the drivers of the forecast are, including the forecaster’sconfidence in the model, recent errors and other relevant information

partic-Step 2

The participants are asked to use their knowledge and market ence to estimate the extent to which the objective forecast for the yieldought to be changed and to write down the factors involved That is,the participants are not asked to make a forecast from scratch but toanchor it to the objective statistical forecast If the team would like toremove anchoring to the statistical forecast, however, individuals areasked to construct their forecast independently of the model-based one

experi-In their example, Makridakis, Wheelwright and Hyndman refer to aform that can be completed to facilitate the process For yield forecasts,this form would contain a wide range of influences on yields Thestatistical model makes use of fundamentals such as rent growth andinterest rates to explain real estate yields, whereas the form containsfields pointing to non-quantifiable factors, such as the momentum andmood in the market, investment demand, liquidity, confidence in realestate, views as to whether the market is mis-priced and other factorsthat the participants may wish to put forward as currently importantinfluences on yields This form is prepared in advance containing allthese influences but, of course, the house view participants can addmore If a form is used and the statistical forecast for yields is 6 percent for next year, for example, the participants can specify a fixedpercentage per factor (strong momentum, hence yields will fall to 5 percent; or, due to strong momentum, yields will be lower than 6 per cent,

or between 5.5 per cent and 6 per cent, or between 5 per cent and 5.5 percent) This depends on how the team would wish to record the forecasts

by the participants All forecasts have similar weight and are recorded

Step 3

The individual forecasts are summarised, tabulated and presented toparticipants, and the discussion begins Some consensus is expected

on the drivers of the forecast of the target variable over the next year

or years In the discussions assessing the weight of the influences, theparticipants’ ranks and functional positions can still play a role andbias the final outcome All in all, this process will result in agreeing theorganisation’s final forecast At the same time, from step 2, there is arecord of what each individual said, so the participants get feedbackthat will help them improve their judgemental forecasts

Real estate forecasting in practice 431

is a simpler interaction in the house view process This informalarrangement makes it more difficult to record judgemental forecasts,however, as the discussion can kick off and participants may make uptheir minds only during the course of the meeting

The outcome of the house view meeting may be point forecasts overthe forecast horizon It may also be a range of forecasts – e.g a yieldbetween 5.5 per cent and 6 per cent The statistical forecast can be taken

as the base forecast around which the house view forecast is made Forexample, assume a statistical forecast for total returns over the nextfive years that averages 8 per cent per annum The house view meetingcan alter the pattern of the model forecasts but, on average, be veryclose to the statistical forecasts Furthermore, point forecasts can becomplemented with a weighted probability of being lower or higher.This weighted probability will reflect judgement

Given the different ways to intervene in and adjust model-based forecasts,

a way forward is illustrated in figure 13.1 The value for 2007 is the actualrent growth value The model-based forecasts for 2008 and 2009 are given

by the plain triangle In all probability these forecasts will not be entirelyaccurate, as the error will incorporate the impact of random events, andthe actual rent growth values for 2008 and 2009 could be either of the twoshaded triangles – that is, the actual rent growth will be higher or lowerthan predicted by the model

Expert judgement can come in two ways to modify this forecast

(1) By weighting additional market information, a probability can be given

as to which direction the actual value will go In the figure, such a ment may suggest that, based on market developments not captured bythe model, there is a greater probability that rent growth will be lower

judge-432 Real Estate Modelling and Forecasting

than that predicted by the model in 2008 but higher in 2009 (as shown

by the black triangles)

(2) The expert intervenes to provide an absolute forecast, shown by thecrosses for 2008 and 2009 in the figure We explained earlier in thechapter how this absolute intervention can take place; it can be arbitrary

or it can utilise previous errors of the model

In any event, this chapter has highlighted two other issues: (i) the wholeprocess should be transparent and (ii) a record should be kept so that theforecasts, of whatever origin, can be evaluated using conventional forecastassessment criteria

13.7 How can we conduct scenario analysis when judgement

is applied?

Scenario analysis is straightforward from a regression model We can obtaindifferent values for the dependent variable by altering the inputs to allowfor contingencies Judgemental intervention does not preclude us from car-rying out scenario analysis Some forms of judgemental mediation make itdifficult to run scenario analysis, however A prerequisite is that the finalforecast is partly model-based For the most part, we can run the scenariousing the statistical model, and we then bring in the judgement we origi-nally applied This is an additional reason to ensure that the judgementalinput is well documented when it is applied to the quantitative forecast.With pure judgemental forecasts, scenario analysis is somewhat blurred

as a process The expert holds a view, and it is not clear how the question

‘What if ?’ can be answered apart from direction The expert can, of course,

give higher or lower probabilities about the outcome based on differentscenarios This is easy when the scenario is based on economic conditions,but if the expert’s forecast utilises information from contacts within theindustry it may be more difficult to work out the scenarios

13.8 Making the forecast process effective

The previous sections have identified factors that will make the tion’s forecast process more efficient when statistical forecasts and judge-ment are combined Bails and Peppers (1993) look into how the gap betweenforecasters and users (internal or external) can be bridged, and discuss theforecaster’s responsibilities and how to get management to use the forecasts.Drawing on Bails and Peppers’ and other studies, a number of suggestionscan be made

organisa-Real estate forecasting in practice 433

(1) Periodic meetings should be held between the preparers and the users

of the forecasts The meetings should involve management and experts

in the forecasting process

(2) The forecaster should explain the nature of forecasting and the problemsinherent in the forecast process What are the limits to forecasting? Whatcan quantitative forecasts not do?

(3) The forecaster should also explain the meaning and the source of theforecast error The aim in both (2) and (3) is to direct the attention of theexperts to the gaps in statistical modelling

(4) The forecaster should understand the user’s objectives Consumers

of forecasts may be more interested in why the forecasts might notmaterialise

(5) The forecaster should be prepared to test ideas put forward by expertseven if these ideas are more ad hoc in nature and lack theory

(6) The usefulness of forecasts is maximised if contingency forecasts areincluded Scenario analysis is always well received

(7) Technical jargon should be kept to a minimum The forecaster needs to

be clear about the techniques used and endeavour not to present themodelling process as a black box

(8) Always incorporate a discussion of historical forecast accuracy and adiscussion of how inaccuracies have been addressed If there is a record

of expert forecasts, the forecaster can, ideally, calculate the followingmetric:

total error= model error + managerial error

The error is decomposed into one portion, which is the model’s sibility, and the residual, which represents a discretionary adjustmentmade by management In this way, all parties gain a perspective on theprimary sources of error

respon-Key concepts

The key terms to be able to define and explain from this chapter are

● forecast mediation ● issues with forecast intervention

● judgemental intervention ● acceptability of intervention

● domain knowledge ● mechanical adjustments

● reasons for intervention ● ‘house view’

● forms of intervention ● intervention and forecast direction

The way forward for real estate

modelling and forecasting

Learning outcomes

In this chapter, you will find a discussion of

● the reasons for the increasing importance of forecasting in realestate markets;

● techniques that are expected to apply increasingly in real estatemodelling;

● formats that forecasting can take for broader purposes; and

● the need to combine top-down with bottom-up forecasting

Real estate modelling and forecasting constitute an area that will see notableadvancements in the future, and progress is likely to be achieved in severalways The methodologies and techniques we have presented in this bookwill be more widely applied in real estate analysis We also expect to seethe employment of more sophisticated approaches in real estate Such tech-niques are already applied in academic work on the real estate market andcould be adopted in practice

There are several reasons why modelling and forecasting work in the realestate field will grow and become a more established practice

● The globalisation of real estate capital and the discovery of new marketswill prompt a closer examination of the data properties and relationships

in these markets Comparisons will be required with more core markets.Investors are interested in establishing possible systematic relationshipsand studying the sensitivities of real estate variables in these markets totheir drivers Investors would also like to know whether these marketsare forecastable

● Greater data availability will facilitate modelling in real estate kets Real estate series are becoming longer, the data are available at

mar-434

Real estate modelling/forecasting: the way forward 435

an increasingly high frequency, and data can now be found in tions that previously had very little data New and expanding real estatedatabases pose challenges to analysts Analysts will be able to test alter-native theories and models with the aim of finding the best forecastingapproach

loca-● Forecasting will also be underpinned by education trends Larger bers of analysts with the appropriate skills enter the industry nowadays,partly as a result of more universities including quantitative modellingstreams in real estate courses These analysts will utilise their skills, andthe emphasis on rigorous forecasting should be stronger The wealth oftechniques applied in other areas of economics and finance will attractthe interest of real estate modellers to assess their applicability in thisfield

num-● There are also external pressures to undertake formal forecasting As thereal estate industry rises to the challenge to be a mainstream asset class,

it should be expected that objective forecasting will be required A acteristic of this market is that it follows economic trends fairly closelyand is more forecastable (the occupier market, at least) than other assetclasses Real estate modellers will have to provide increasing evidencefor it

char-● There will be more sophisticated demands in real estate modelling thatcan be addressed only by econometric treatment, such as forecasts andsimulations for the derivatives market We describe such demands later

in this chapter

Regression analysis will remain the backbone of modelling work and willcontinue to provide the basis for real estate forecasts The use of regressionanalysis rather than more sophisticated methods reflects the fact that, inmany markets, there is a short history of data and, in several instances,the series are available only at an annual frequency In markets and sectorswith more complete databases, multi-equation specifications will offer agood alternative to single-equation regression models for forecasting andsimulations These two forms have traditionally been the most widely usedforecasting techniques in real estate practice The concepts behind thesemodels are easy to explain to the users of the forecasts, and the process ofperforming scenario analysis is very straightforward These frameworks, inparticular single-equation regression models, are often taken to provide thebenchmark forecast

There is little doubt, however, that the other techniques we have sented and explained in this book will be used Given the suitability of VARs

pre-436 Real Estate Modelling and Forecasting

for forecasting, these models will present a useful alternative to researchers,especially in markets with good data availability They will provide a use-ful framework for forecasting quarterly and monthly series – for exam-ple, indices used for the derivatives market ARIMA methodologies are alsoappealing for short-term prediction in particular, and for producing naiveforecasts Given the availability of software, such models can be constructedquickly for forecasting purposes Cointegration is undoubtedly gainingground as a technique for the analysis of real estate markets More and morerelationships are examined within a long-run equilibrium framework, anappealing theoretical concept, whereas the information additional to short-term adjustments from the error correction term cannot be ignored Realestate researchers will be investigating the gains emanating from adoptingcointegration analysis for forecasting

One of the effects of globalisation in real estate has been the need tostudy new but data-constrained markets A framework that researchers will

increasingly be employing is panel data analysis This represents a whole new

area in applied real estate modelling When time series observations arelimited – e.g when we have end-of-year yield data for six years in a location –

it is worth investigating whether we can combine this information withsimilar series from other locations – that is, to pool the data Assumingthat we have, say, six years of data in ten other locations, pooling the datawill give us around sixty observations We can then run a panel model andobtain coefficients that will be used to forecast across the locations

Pools of data obviously contain more information than pure time series orcross-sectional samples, giving more degrees of freedom, permitting moreefficient estimation and allowing researchers to address a wider range ofproblems The use of a panel can enable them to detect additional features ofthe data relative to the use of pure time series or cross-sectional samples, andtherefore to study in more detail the adjustment process of the dependentvariable in response to changes in the values of the independent variables

In some instances, it is permissible to pool the time series and cross-sectionalelements of the data into a single column of observations for each variable;

otherwise, either a fixed effects or a random effects model must be used Assume

we estimate a model for yields Fixed effects will help us to control foromitted variables or effects between markets that are constant over time(reflecting certain local market characteristics) In certain markets, however,the impact of these variables may vary with time, in which case the timefixed effects model, or possibly a random effects model, should be chosen

A comprehensive treatment of panel data estimation techniques and theirapplication is given by Baltagi (2008); an accessible discussion and examplesfrom finance are presented by Brooks (2008, ch 10)

Real estate modelling/forecasting: the way forward 437

Future real estate research will focus on identifying early signals in boththe occupier and investment markets The real estate industry is movingtowards more timely analysis This research will take a number of forms

We have seen a good volume of work on the relationship between direct andsecuritised real estate The argument is that, due to the frequent trading

of the latter, prices adjust more quickly than in the direct market Thesmoothness of the direct real estate market data is partly the reason for theslow adjustments in this market As a consequence, the securitised marketcan be used for price discovery in the direct market Given the increasingnumber of REITs around the globe, particularly in markets in which directreal estate market data are opaque, REIT market signals will be studiedclosely and included in the forecast process

Research on early signals will and should focus on leading indicators.Leading indicators are used to capture changes in direction and turningpoints There is a significant amount of research being conducted in eco-nomics and finance that quantitative analysts will naturally be applying toreal estate Relating to leading indicators is the topic of predicting turningpoints Again, there is a large body of work in economics on turning points,and we should expect researchers to utilise the insights of this literature forapplication to the real estate market

The econometric models we have examined in this book can be mented with leading indicators For example, a model of rents in the UnitedStates can include consumer expectations and building permits, which areconsidered leading indicators of the US economy by the Conference Board

aug-A model of rents with or without leading indicators will attempt to dict turning points through the predictions of future values of rents Theprediction of turning points is therefore a by-product of the point forecasts

pre-we make for real estate variables There is another family of econometricmodels, the specific objective of which is to identify forthcoming turningpoints and establish probabilities for such occurrences The difference withstructural models is that the forecasts they make for, say, rents are based onthe ability of these models to track past movements in rents In the secondcategory of models, the prediction of turning points reflects their ability topredict turning points in the past

One such class of specifications is what are known as limited dependent able models, in which the dependent variables can take only the values zero

vari-or one This categvari-ory of models includes probit and logit models Fvari-or ple, we can construct a variable that specifically isolates negative returns

exam-for Tokyo, say where the values that the returns can take are y i = 1 if total

returns are negative and y i = 0 otherwise We can use leading indicators toassess the likelihood of a turning point in Tokyo office total returns and also

438 Real Estate Modelling and Forecasting

evaluate the past success of the chosen models These models are expected

to show rising probabilities when turning points in total returns are about

to occur A relevant study in the real estate market is that by Krystalogianni,Matysiak and Tsolacos (2004)

The leading indicators used in models to predict the change in direction

or turning points could, of course, be real estate variables themselves Forexample, series of active demand, which are registered queries by occupiersfor space, could be seen as a harbinger of take-up or absorption Alterna-tively, surveys of expectations in the real estate market could also provideearly signals In any event, the success of these variables should be assessedwithin these different classes of models

The key challenge for models focusing explicitly on turning points is thefrequency and history of data If these models are used to predict severedownturns in the real estate market, there are only three or four majordownturns that can be used to train the models Of course, the turning pointcan be defined more loosely, such as when returns accelerate or decline –and not necessarily when they become negative or positive The smoothness

of the real estate data can be another issue in the application of probit orlogit models

As global markets become more interlinked, we would expect futureresearch to focus on the transmission of shocks from one market to another.This work will replicate research in the broader capital markets – for exam-ple the bond market – in which the transmission of volatility between mar-kets is examined Again, through the VAR and VECM techniques we studied

in this book, we can trace such linkages – e.g through impulse responses.There are of course other methodologies, such as the so-called multivariateGARCH (generalised autoregressive conditional heteroscedasticity) models,which are geared towards the study of volatility and volatility transmission.Existing work on this topic in real estate includes the study by Cotter andStevenson (2007), examining whether bond market volatility transmits toREIT volatility, and the study by Wilson and Zurbruegg (2004), who considerhow contagious the Asian currency crisis was in 1998 and the impact onreal estate in the region

Transmission can be studied to address questions such as the following

● Do returns in Tokyo offices become negative following negative returns inNew York, how severe is the impact and after how long does it dissipate?

● What is the probability of negative growth in office rents in Hong Kong

if Hong Kong REIT prices fall?

● What is the impact of NCREIF and IPD capital value changes on eachother?

Real estate modelling/forecasting: the way forward 439

In real estate, predictions are often expressed as point forecasts rally, one should expect that the future value for rent growth will almost

Natu-certainly be different from the point forecast The study of uncertainty

sur-rounding a forecast is appealing to investors, particularly in downturns.Prediction intervals can be used to show the possible size of the future errorand to characterise the amount of uncertainty In this way, the uncertaintyabout the model and the possible impact of a changing environment can

be depicted This interval forecast consists of upper and lower forecast

lim-its and future values are expected to fall within these boundaries with aprescribed probability At different probabilities, these boundaries can bewider or narrower around the central forecast Interval forecasts and, moregenerally, estimates of the probabilities of different outcomes are valuable

to underwriters, rating agencies and risk managers

Forecast evaluation is an area that has received little attention in realestate so far, but that will change The credibility of models is heightened

if we can demonstrate their accuracy in tracking the real estate variable weseek to explain and forecast It is also important to assess past problems,

to explain what went wrong and to determine whether this could havebeen caused by incorrect inputs There is more focus on demonstrating thevalidity of models, and users would like to know what the models do notaccount for Trained producers of forecasts will be adopting forecast eval-uation techniques Based on this expectation, we have devoted a separatechapter to this subject area

Finally, there is no doubt that judgemental forecasting will remain a ture in real estate prediction The success and acceptance of model-based asopposed to judgemental forecasts will be evaluated, as we have discussed

fea-in the book, by an assessment of their out-of-sample forecast accuracy casters and experts will be working together more closely so that they canbetter understand how to combine their information

Fore-Bottom-up forecasting has always been the main approach in real estate

markets Asset managers assess the investment with regard to the ities of the building, the tenant characteristics and other attributes Anumber of techniques are used to establish whether the building is fairlypriced We expect to see more work on bottom-up forecasting and a greater

qual-effort to combine it with top-down forecasting With a greater

availabil-ity of data we should see more formal forecast techniques being applied

to price the risk factors at the building level and predict returns at theasset level As we will increasingly move into situations of scenario fore-casting and stress-testing our estimates, both top-down and bottom-upapproaches to forecasting will be valuable to carry out these tasks at the assetlevel