Forecasting Stock Market Strength Using Federal Reserve Data

Posted on May 9, 2014 by

5



It is clear, in hindsight, that investors would have done better in 2008 if they had known to move from stocks to fixed income at the beginning of 2008. And similarly, if we could anticipate future bull markets we’d like to be able to shift assets into equities beforehand. Is it possible to anticipate these significant market moves?

Note: This article describes a macro economic model we’ve been working on at Lucena Research. We developed it as a way to inform some of our investing strategies — As one example: For a large cap stock & bond strategy we’d like to know whether we ought to be emphasizing stocks or bonds at any particular moment.

How can we predict market strength or weakness?

Short-term market timing is challenging, and may not be appropriate for conservative portfolios. However, we believe that macro economic indicators have strong longer-term predictive capability, and that it makes sense to pay attention to them.

We’d like a quantitative tool to help us anticipate strength or weakness. We began a search for clues in government data. In particular, the Federal Reserve retains a wealth of historical measures of our economy over many decades. After investing several person-months scouring the available Federal Reserve data, we identified 10 factors as the most useful in predicting market strength or weakness. A few of them include:

  • Continued Claims on Insured Unemployment
  • Delinquency Rate on Business Loans
  • Civilian Employment-Population Ratio
  • Federal Debt: Total Public Debt as Percent of GDP

An ensemble of economic “experts” 

Our intellectual origin is statistical machine learning, and one of the most important developments in that field over the last decade or so is the idea of ensembles of experts.  Essentially, researchers have discovered that we can make predictions much more effectively if we consult many different experts than if we consult just one or two.  We’ve carried that lesson forward here to create a group of simple “experts” to help with our market forecast.

We developed a rule of thumb (or heuristic) for each of our 10 factors.  We treat each heuristic as an individual “expert.”  If the heuristic predicts strength, the expert votes +1, otherwise it votes 0.  As an example, a rapid rise in the Public Debt to GDP ratio can be predictive of market weakness.  So the corresponding heuristic might vote 0 (or “no”)  in the case of rising debt.

In our testing it turns out that some of these experts have been more effective than others. So we give them different weights. Those experts who have provided better predictions in the past get a stronger weighting than those with weaker records. We sum these votes each day to create a combined outlook for the future (see figure).

Observe that the model anticipates the significant downturns in late 1990, 2000 and 2008. It also anticipates all of the significant bull markets since 1989. There are certainly other factors that may also inform overall market prediction and we anticipate adding more factors to our model as we discover them.

 

Lucena's macro economic model for market strength.

Lucena’s macro economic model for market strength (top, blue) and S&P 500 performance (bottom, red). The macro model’s value ranges from 0 (weakness) to 10 (strength).

 

Use in a simple strategy simulation

One way to test this model is to see how well it works with an investment strategy. For this test, our strategy is simple:

  • If the macro model value is greater than 0.5 buy the S&P 500, otherwise
  • If the macro model value is less than 0.5 buy short term bonds

The results of this backtest are shown in the figure below. The simulation runs from June 1988 until early May 2014. The strategy returns 991% compared to 585% for the market. The key difference between the two is that the strategy exits the market prior to the major downturns beginning in 2000 and 2008.  The apparent underperformance from 1991 to 2000 is due to the strategy remaining out of the market a bit too long after the 1990 downturn.

 

Using the macro model in a simple strategy (green), compared to the market overall (purple).

Using the macro model in a simple strategy (green), compared to the market overall (purple).

 

Not a standalone strategy

This quantitative tool isn’t intended to be a strategy all by itself.  It is intended to inform a strategy. For instance, it probably isn’t best to be only “in” or “out” of the market. Instead one should overweight one set of investments or another as the forecast shifts.

How to get it

Lucena’s macro economic model is available to our clients via the QuantDesk(r) platform, and also as part of our BlackDog strategy. Please contact us via email at info@lucenaresearch.com for more information.

by Tucker Balch, Ph.D., and Scott Strong

Advertisements