In the Nelson Bakewell/South Bank University lecture, Paul McNamara argues that the sooner surveyors are able to understand the work of the property forecaster, the better.
Dr Paul McNamara is Property Research Manager at Prudential Portfolio Managers. This paper was originally given as the 1994 Nelson Bakewell/South Bank University Annual Estate Management Lecture.
There has been a substantial growth in producing forecasts for the property market during the past decade. I can think of a number of chartered surveying firms – including Hillier Parker, Jones Lang Wootton, Richard Ellis and St Quintin – that produce forecasts Several independent organisations produce forecasts annually or quarterly: Property Market Analysis, Real Estate Strategy, Barber White and the Investment Property Databank itself. Firms such as my own regularly produce forecasts on different aspects of the UK market. The jargon of the “quant” – the quantitative analyst – is beginning to enter the everyday language of the surveying profession. With the current cynicism among many about the quality of advice being provided by chartered surveyors, and the rapid expansion and increased visibility of research services, the traditional surveying profession could be forgiven for worrying where this invasion of “PhDs”, spouting off about their best r-squared, time-lagged, autoregressive, log-transformed econometrically derived equation, is leading.
My theme for tonight is that, when thinking about property forecasting, there is, in the words of President Roosevelt, nothing to fear but fear itself. My argument is twofold. First, that the property market forecaster is essentially engaged in making explicit what has long been implicit in the investment advice given by surveyors. Second, that the skills of the surveyor and the property market forecaster are entirely complementary. To this end, I shall deal in some detail about the forecasting process as it operates at Prudential Portfolio Managers.
To illustrate the point about forecasters making explicit the implicit, I remember an exercise carried out in one of the national property journals in summer 1990. Six surveyors were contacted to give their views about how the City office markets would develop over the next five years.
The results were fascinating because they revealed the variety of “models” that were being carried around in the heads of these surveyors – and, one presumes, they would influence the way in which they would value property in the City of London. The surveyors were asked to give estimates of current yields and rents and provide 1995 forecasts for a hypothetical quantum of space in three locations (Holborn Viaduct, the Bank of England site and Minories) which traversed the City of London. The figures for a 150,000-sq ft refurbishment at Holborn Viaduct are presented in Table 1.
The level of variation is not at question here – I feel sure that they are no more varied than those produced by equity analysts in making predictions about company prospects. However, what is interesting is the variation in the reasoning between the various market pundits. As can be seen, some would value the property differently because of varying views on the appropriateness and attractions of refurbishing the building (Surveyors A and B).
It is less clear whether these surveyors perceive any material change in the relative attractiveness of this location. By contrast, surveyors D and E perceive that the nature of this area as an office location may be transformed through either the Holborn market filling in between the City and the West End to make the central London office market fully contiguous or the success of a scheme at Paternoster Square.
Some of the ideas were clearly products of their time – no reason for criticism there. The main point to draw from Table 1 is that once implicit models about the future of property markets which are held in the heads of experienced surveyors are made explicit then we can see substantial variety.
Surveyors often make decisions based substantially on models of how markets will develop that are “experientially” based. These are powerful models indeed. People do not work 15 and 20 years in an industry and learn nothing! However, there is a continuing need to tease out and test (and re-test) these models over time.
My concern is that, without such exercises as the one presented above which tease out underlying logic, the reasoning behind many of the strongly held views within the property market is rarely capable of critical review. Much is made of the surveyor’s “gut feel” about markets, sectors and opportunities. However, if the views held or expressed remain largely unarticulated, except for the conclusion, then discussion and debate are diminished. How can recipients of the view be satisfied that all the relevant information has been considered, that the logic leading to the conclusion is correct, that the views held on related issues are consistent: that two surveyors with the same conclusion hold it for the same reasons, and that the assumptions underpinning the view are plausible.
In most cases, the experientially based model will serve to determine an appropriate course of action, but I would suspect that “gut feel”‘ is at its weakest in periods of market change.
Forecasters are perennially engaged in the sort of process described in Table 1, except they do it in a more explicit way, often drawing on different data sources and in a manner that allows assumptions to be critically reviewed. Their databanks not only enable models of past, current and future market behaviour to be built but they also facilitate the construction of rational expectations. For example, when shop yields in Norwich reach 3.5%, suggesting, say, a required rate of net rental growth (ie accounting for depreciation) of 6% nominal, in perpetuity, the historical record can be objectively reviewed to see when, why and under what conditions that rate of rental growth had been previously achieved. This allows a probability to be placed on the likelihood of such an assumption being borne out and, hence, being tenable as a central element of a valuation. Surveyors and forecasters are often doing similar tasks – but differently.
Each quarter, led by our property economist Paul Mitchell, my team is charged with the responsibility of producing forecasts of the likely performance of the UK property market and its constituent parts. Currently, this involves the production of annualised forecasts for a sequence of five periods of rental growth and expected total returns for the three main sectors of the market in each of the 13 economic regions of the UK. In this respect, what we produce in-house mirrors the services provided by other forecasting agencies in the market-place.
We choose to do the forecasting process in-house because it provides a greater level of control over the process than is obtainable by simply letting others run their forecasting models, with our economic forecasts, for us.
It facilitates a close interaction with the general economic forecasting process led by our UK economist and, of critical importance, our practising surveyors. These property forecasts serve a number of inter-related purposes. First, and most important, they are submitted, in aggregated form, to an Asset Allocation Committee, alongside forecasts for other investment media. This asset allocation process then determines the level of cash flow into and out of property portfolios.
Clearly, property decisions are taken for a longer period than the monthly “touches on the tiller” used for the mass of other, more liquid, assets. Once the decision has been made to inject more capital into or out of the relevant property portfolios, the sector and regionalised forecasts are then used to guide the process of determining property fund strategy. With a knowledge of the structure of our funds and the structure of their respective bench-marks it is then relatively straightforward to establish how our funds will perform if they (and their bench-marks) remain inactive.
However, with forecasts of return and a knowledge of how our funds are configured in respect of their bench-marks, it is possible to suggest courses of action which will (a) improve absolute and relative returns to the fund and (b), possibly, reduce the risks to which the fund is exposed. Clearly, if one is underweight relative to a bench-mark in a given sector-region that is going to perform well, then buying stock in that market will improve absolute returns, because the market is expected to perform better than other markets.
It will improve the relativity of returns against competitors (in this case, making sure they don’t steal too much of a march, but it could equally be by making a stronger play than one’s competitors in a good market) and would reduce an underexposure to the bench-mark, thereby reducing the risk that the bench-mark might experience very much stronger performance than your forecasts might predict.
Finally, though I cannot say too much about what is a proprietary approach to valuing property assets, the sector-regional forecasts (plus other research-based elements) form a core of the valuation process from which PPM works.
In this way, the forecasts we produce assist in determining the flow of capital in those property funds over which we have direct control. They assist in determining whether PPM is getting good value when buying or continuing to receive good value from assets already held in portfolios. In case this sounds somewhat mechanical, let me emphasise that this process is not deterministic. We are always aware that, for whatever reasons, good-value properties can be found in poor-value markets. It is an investment process which we believe is coherent from the macro to the micro level and it has property market forecasting at its heart.
What makes for good forecasting? The essence of good forecasting, in my mind, is basing one’s models of how rents vary over time firmly on a good understanding of how economic and social processes affect the demand for, and supply of, property. There are two ways in which one could go about establishing relationships between rents, demand and supply. One could, first, theorise on what factors drive rents and then collect relevant information upon which to build one’s models or equations.
A second approach is not to presume any prior knowledge and simply put data on a large number of variables into the computer and let the computer tell you which variables best explain, say, rental growth in a given land-use sector. Most likely, there will be a necessity to do a little of both. For example, taking the retail sector, the basic theoretical stance might be something very straightforward, such as rents for retail property are a function of the “productivity” of retail space – by which is meant, the level of sales that are effected through a given quantum of space. When one then moves to produce forecasts of retail rents does one make use of retail sales forecasts, retailers’ profitability forecasts or forecasts for consumers’ expenditure to estimate the demand side of the equation?
Similarly, does one use forecasts of construction tenders or local authority planning applications registers to forecast supply? Hence, in practice, there is both theory and there is an element of searching for the relevant data which produce the best estimates. There are a number of things to be noted at this juncture. First[1] , the root of most forecasting used in the property world is regression analysis of some form, when one estimates and/or predicts the value of one variable by reference to other variables.
As such, it is feasible to find accurate but nonsensical relationships between two variables – increases in the 1970s in UK unemployment and the consumption of French Beaujolais is one that I have heard of. However, to retain confidence in a forecasting model or equation, we want to feel that the relationships being described are valid rather than simply accurate. Second, we would want to feel confident that the relationship defined by the model was reasonably stable over time and reasonably transferable from one situation to another. Given the “fundamentalist” nature of the enterprise I have described, it would undermine confidence if one’s model was good at explaining what happened in the 1960s and 1980s, but not the 1970s. How sure are we that it has captured the fundamental processes that would have been at play throughout that period if it is unable to explain a key part of it?
Similarly, how confident would one feel about a model that accurately forecast rental growth in the Manchester office market but was hopeless at predicting that in Leeds? One might expect some minor differences in the finer details of how the model is calibrated – but one would not feel happy if variables that had explanatory power in one instance had none in other, ostensibly similar, situations. Clearly, the statistical evidence of good relationships between meaningful variables is a central tenet of good forecasting but, given the complexity of the real world, it would be both foolish and arrogant to believe that the forecasting model could do all the work for us.
To help us in providing a better forecasting service for our funds we, first, make extensive use of “scenarios”. If we believe that there is more than one materially different economic or property market outcome, using these different impacts possible, we explicitly use our model to forecast those outcomes using those different inputs. We can then reflect on the probability of each “future story” actually unfolding. A good example now would be plotting out the effect on performance under different interest rate regimes by the year-end.
Second, we also make extensive use of the (property) market-facing skills available to us within our property division. Our surveyors have, first hand, daily interaction with the market. We make use of this by involving them in the forecasting process. Their appreciation of the tenant and investment market is used by the research team in qualitatively reviewing what the quantitative modelling exercise has suggested. Hence, we believe that econometric models serve to point objectively towards (and put bounds of credibility around) a likely level of, say, rental growth. The qualitative review finishes that process for any given scenario. Mechanical forecasts are to be avoided, as is “data mining” for its own sake. One’s model should be plausible, replicable and based on a good understanding of relevant economic and social processes.
Before assessing how forecasters are to be judged, it is worth briefly highlighting some of the issues that currently face forecasters and, indeed, portfolio constructors. On a practical level first, it is important to note that much of the forecasting work which is done in the UK is “data-constrained”, in that is making do with less-than-adequate information. More information offers the prospect of better forecasts being produced.
The discontinuance of floorspace statistics by central government, the patchy quality of local authority data on floorspace in the development pipeline, the abandonment of Censuses of Distribution and the almost ad hoc nature of the Census of Employment all conspire to reduce the ability to forecast property markets well. Yet a modicum of money spent here could well help to avoid the mass of underemployed capital ending up in empty office buildings across the length and breadth of the UK. Second, and much more conceptually (albeit rooted in the issue of forecasting being data-constrained), the reason why most property market forecasts available in the UK at the moment are based on either sectors or sector-regions is because the data to produce forecasts comes in those units, not because the units make intrinsic sense as meaningful divisions of the property market. When one talks about forecasting a “property market”, or constructing a property portfolio by distributing investments across various “property markets”, what do we mean? I don’t think any of us would define the East Anglian industrial market as a functional, meaningful entity – but yet we produce forecasts for it.
Forecasts should be produced for a comprehensive range of meaningful entities – and this is why most forecasting teams are seeking to move down to the town level in their forecasting process. In part, this still represents a data-constrained solution – for example, perhaps the most meaningful divisions in the property market are not spatial; perhaps they refer to age and size of buildings. However, there are no data series outside IPD to test whether this is the case or which would provide time-series of data to allow rents, age and size to be statistically modelled.
Towns are seen by most analysts as more meaningful than regions and are now being addressed more centrally by property market forecasters. It is interesting to note that this development will mean that instead of producing 3 x 13 forecasts every quarter, something like 3 x 50 (for the 50 biggest property markets) forecasts will need to be produced. It also means that the forecasters will have to come down from the level of statistical abstraction that working at the regional scale affords them in order to interact much more closely with practising surveyors who know local markets well – but from an experiential rather than a high-level political economic perspective. Given these sorts of major issues, it is clear to see that property market forecasting is very much in its infancy.
I have a list of at least 30 different projects which we would like to carry out to improve our forecasting ability and, as I am sure you will agree, the more one learns, the less one seems to know.
How should forecasters be judged? Those unfamiliar with, or wary of, the forecasting process should resist the immediate feeling of schadenfreude when the out-turn in a given year was that rents grew, say, by 10% although the research team had predicted 15%. Before being over-critical, the reviewer should ask a few preliminary questions. They should seek to explore, first, were there any good reasons for the forecasts being blown off course over the year? A good example here is that property forecasts are centrally based on economic ones.
If there have been economic shocks, for example as a result of the Gulf war, they will have affected the basic inputs for the property forecasting model. That said, it is perfectly legitimate to explore the accuracy of the property model after these shocks had been accounted for. There is also a need to determine whether the forecast had given the investment process the right message. If the basic message was to buy or sell in a given market, and the out-turn more than vindicated this, then the purpose of the forecasting process was served.
However, if the forecasts led one to do things one should not have, or not do things one should have, there is legitimate room for concern and criticism. Hence, it is the message rather than the absolute “point accuracy” of the forecast that is important. Finally, the reviewer should ask whether better advice could have been made available elsewhere. How did the forecaster perform against the reviewers or the general market’s intuition? How did the forecaster perform against other forecasting services? Clearly, such judgments should be done honestly and over a reasonable length of time. Hasty and post hoc rationalisations have no place in this process.
How may surveyors and forecasters coexist in the future? To pick up our earlier theme, I see no problems with a future, mutually beneficial, coexistence of surveyors and forecasters. Both groups possess complementary sets of skills and information. Both have distinct rather than coincident roles to play in the investment process.
Forecasters will tend to inhabit a plane of analysis geared to understanding how macro processes may evolve and trigger the ground swell of movement in national and subnational property markets. However, this Olympian perspective does not equip them to take an “intellectual” high-ground in what is essentially a partnership with investment professionals who inhabit a plane of analysis which better understands the nuance of market behaviour, market context and the infinite variety provided by buildings. That said, there needs to be a far better understanding and appreciation of each other’s contribution. This need not be passive acceptance of each other’s views on any given issue. A surveyor has every right to have the logic of any given forecast laid out for review – or to challenge the basic assumptions underpinning a forecast. Similarly, the forecaster may well critically review the true importance of micro processes outlined by surveyors as affecting a given market. In our company, with the exception of tracking major investments such as shopping centres, we have a very clear view that regional analysis is the province of the researcher; anything to do with buildings is the province of the surveyor – and town-level analysis belongs to both. By better understanding what each other does, one would hope that a level of trust between the two groups will emerge.
However, that trust is of little use if the forecaster is really only listened to when he or she confirms the investment manager’s own prejudices and otherwise is ignored. With trust comes the requirement for influence. The forecaster has a right to be heard even when the message is unpalatable.
So, in conclusion, I would repeat, there is nothing to fear but fear itself. Forecasting is a complementary skill to that of traditional surveying and will help property investment advisers in performing their duties in a more disciplined and reviewable mariner. It is the way of the future and the sooner surveyors are able to understand the work of the forecaster, the better. This has clear implications for higher education courses. Moribund statistics options in undergraduate surveying courses should be revitalised and refocused on helping the practising surveyor to understand and review critically the work of the “property economist” with whom he or she will likely be partnered in later life.
[1] For a more substantial discussion of these issues read Field and MacGregor (1987), Forecasting techniques for urban and regional planning Hutchinson, London.
