Not the usual Statistical Approach.
We took a different approach. We wanted to know what really were the underlying drivers for the construction industry because our client had a lot of customers there and big development costs being dedicated to it. We recognised that for the next two decades, all the people who would be buying houses had already been born. We took account of the age specific activity rates, such as ages of marriage, rate of births per family, number of divorces and ages of divorcees and so on.
It turns out that many of these parameters change very slowly, society develops its norms of behaviour much more slowly than governments can change the rules. We created a model of the housing stock and the demographic distributions in the population and looked at how the resulting rate of household formation, which includes all the new singles as well as nuclear families, would create a demand for places to live. (dwellings including apartments (flats) as well as single family homes).

Good match to reality
From 1945 on to about 1970, our intrinsic demand model tracked the observed rate of creation of dwellings quite well. There was a glitch period, with far fewer houses built than our model predicted, but then we discovered that it had been a period of government imposed building restrictions (for material shortages). After the glitch, the real world caught up with our model again.

Projections into the Future
The next question was "what of the future?" The government sponsored statistical model said more of the same. When the house building market turned down one year, the trend line automatically adapted to droop a bit. Our model gave a different view; it said the bottom would drop out of the house building market over the next three years or so because the demographic drivers had gone. Indeed they had. In the next two years, just as we predicted, the market collapsed. The statistical projection was no help.

Other Experience
My experience with problems over these time scales is that people rarely expand their time horizons far enough to see the whole picture. Even if they try, they tend to miss interaction effects and think more in terms of extrapolating, sometimes with statistical models, from the recent past. My (MJMcC) observation is that the demographics and the compelling force of reality can be quantified to make a big improvement in perception. You only have to contact me to get help!