Sunday 26 November 2023: AI Predicts Timing of next House Price Crash
Case-Shiller Real Home Price
This week, we are going to use a Neural Network to forecast the S&P CoreLogic Case-Shiller U.S. National Real Home Price NSA Index. It is a composite of single-family home price indices for the nine U.S. Census divisions and is calculated monthly. It is included in the S&P CoreLogic Case-Shiller Home Price Index Series which seeks to measure changes in the total value of all existing single-family housing stock.
Data for the series since 1890 can be downloaded from Yale University’s Econ page here.
What is a Neural Network?
According to Investopedia (the most jargon-lite source in this instance), a neural network is a series of algorithms that endeavors to recognize underlying relationships in a set of data through a process that mimics the way the human brain operates. In this sense, neural networks refer to systems of neurons, either organic or artificial in nature.
Neural networks can adapt to changing input; so the network generates the best possible result without needing to redesign the output criteria. The concept of neural networks, which has its roots in artificial intelligence, is swiftly gaining popularity in the development of trading systems.
Neural networks, in the world of finance, assist in the development of such processes as time-series forecasting, algorithmic trading, securities classification, credit risk modeling, and constructing proprietary indicators and price derivatives.
A neural network works similarly to the human brain’s neural network. A “neuron” in a neural network is a mathematical function that collects and classifies information according to a specific architecture. The network bears a strong resemblance to statistical methods such as curve fitting and regression analysis.
For our purposes, a neural network is a Big Fancy Squiggle-Fitting Machine, given that we are working with detrended price oscillators i.e. complex, non-linear functions.
1. Output
Firstly, the first 60 years or so of the data is available only as an annual series, and then monthly thereafter. I have used linear interpolation for the annual series in order to have sufficient data points for our exercise.
Secondly, I have placed a Learning Border at the beginning of 1991, thus giving us a 3:1 train-test split. In other words, we will use data until year-end 1990 to train our neural network and test it on the last ~33 years (out of sample data).
Finally, the lower panel shows our actual target variable (aka output) for the model. That is the 100-month Relative Price Oscillator i.e. (price - 100dMA) / 100dMA. It is this detrended oscillator that the neural net will attempt to fit in the training interval, and forecast in the testing interval. We will be looking for:
A positive correlation between the forecast curve and the test data
Close alignment of turning points aka cycle beats.
2. Inputs
We are going to use a bunch of natural cycles that are in close alignment with the major peaks of the spectrum.
3. Results
The result should not come as a shock. The training interval shows that the NN (red line in the lower panel) has identified The Land Cycle aka The Real Estate Cycle aka The Boom and Bust Cycle. There is an argument to be made for the Juglar Cycle as well, but since it has a period roughly half of the Land Cycle we only keep it in mind to identify the termination of mid-cycle slowdown which seems to accompany most instances of the full Land Cycle.
The neural net called the bottom of the early 90s bust and nailed both the peak and bottom of the housing boom/bust of the 2000s.