25 March, 2019

History shows that forecasting property markets, like all markets, is highly inexact science. This blog examines where the models are most accurate, and where they have difficulty.

The IPF Consensus Forecast for 2018 was recently published. The mean forecast for 2018 is 4.6%, with the range of predictions varying from 0.5% to 7.2%. Commentators and strategists across the industry look to this as a useful guide to performance, but history shows that this may not be wise.

Figure 1 shows the accuracy of the IPF Consensus Forecast over time. It compares the forecast of each year, made at the start of that year, with the total return subsequently delivered (e.g. the IPF Consensus Forecast of 2017 total return released in February 2017 is benchmarked against the eventual IPD total return for 2017). The chart shows that even when considering the whole range of predictions, the IPF Consensus Forecast has been accurate less than half the time. In 11 years out of 19, the total return delivered has been outside the range of what forecasters deemed possible.

explaining-forecasting-error-chart-1

There is a clear source of this inaccuracy. Figure 2 compares forecasting error and capital growth not arising from rental value growth – in other words the combination of yield movement and residual. The correlation between the two series is very strong (>0.90), showing that real estate forecasts have been very unsuccessful at anticipating the behaviour of capital markets. Forecasts have, on the other hand, generally been good at predicting the underlying “real” economy, and what will happen to rents.

explaining-forecasting-error-chart-2

Both patterns are also observable when looking at forecasts made over five years; Figures 3 and 4 show that five year forecasts have been accurate in only three cases out of nine, and that the relationship between forecasting error and yield movement is again extremely strong.

explaining-forecasting-error-chart-3

explaining-forecasting-error-chart-4

While this might seem disappointing, it should be noted that the record is improving – five out of the last eight one year forecasts and three out of the last five year forecasts have been correct. This perhaps reflects improvement in understanding and forecasting yield movement. Predictions could well have been becoming more accurate as forecasters have increasingly grasped that while rents might be priced locally, yields are determined globally. It is likely that models have been improved in recent years to more accurately capture the impact of differences in pricing and relativities relative to local and global interest rates, and the future direction of benchmark rates.

For more on this topic, including suggested strategies on mitigating the impact of forecasting error on portfolio performance, please read our companion blog.