If you pay attention to the financial market news you may have noticed a lot of attention being focused on the slowing US/Global economy and the implications it has for financial markets. Just do a search on ‘slowing global PMI’ and watch the hours waste away. Basically, the US/Global economy is slowing which means recession is right around the corner which means financial markets will tank. That seems to be the predominant bear case now, or one of the many. There is some merit to this argument. The worst market downturns occur during recessions. The trick is that you need to know that before the recessions actually happen. In this post, I’ll point you to some research in this area then focus on just one indicator that does a decent job of forecasting recessions and how it can potentially be used as a market timing indicator on its own.

To try and predict recessions there are all kinds of metric and techniques used (ECRI, Conference board indicators, etc). You can spend many many hours looking at all of these and their histories. Believe me. Me and an investor friend have spent tons of hours looking at and studying these. And the history of indicators predicting recessions is mixed to say the least. But I won’t bore you with that here. Instead, if you’re interested, you should read this by Philosophical Economics (which I’ll call PhiloEcon) and some of the linked posts in that piece. There is some incredible work and insight in the post (pretty much anything he/she writes is worth your time). Turns out that historically, the change in the trend in unemployment rate has been a pretty good indicator of recessions. It has also been decent at signaling when the economy has come out of a recession. Below is the key chart.

Screen Shot 2016-05-01 at 9.06.42 AM

Not bad. When the unemployment rate crosses above the 12 month moving average to the upside a recession is likely coming, when it crosses below the 12 month moving average the economy is out of the recession. Can this be used to time the stock market? And does it work better than other market timing indicator such as the popular 200 day simple moving average of prices? Basically, yes. You can read through the post and see how using the unemployment rate improves returns and risk over buy and hold and a trend following system. As usual, I wanted to run some numbers myself. Let’s take a look at that.

I first wanted to see how the unemployment rate indicator (UI from now on) performed on its own versus buy and hold and other trend indicators, specifically the 200 day SMA and 12 month absolute returns. I also wanted to use real investable products, including fees. I looked at returns going back to the beginning of 1999 through April 26, 2016 for the S&P500 ETF (SPY), which fortunately started in 1993. This time period encompasses two of the biggest market downturns in history. I compared buy and holding the SPY versus using the 200 day SMA, 12 month total return, and the UI to exit and enter the market. When the timing systems are out of the market they are not invested, i.e. 0% cash return. Below are the results.

Screen Shot 2016-05-01 at 9.26.04 AM

Very impressive. This simple indicator delivered returns 3.4% per year greater than buy and hold and more than doubled risk adjusted returns. It also beat both other timing systems by a long shot. In addition the simple UI system produced fewer false positives and traded a lot less. Definitely worthy of consideration. You can probably see where I’ll be going next with this. In some following posts I’ll look at adding a risk free asset to the mix during times of risk-off, combining the UI with other indicators (which is what PhiloEcon has done in the GTT system), and adding some global risk assets to the mix. To give you a preview they are all better than what I’ve shown here.

Finally, before I end this post, what is the unemployment indicator saying right now. Does it support the bear case I noted in the opening paragraph. No it doesn’t. The current unemployment rate is 5.0% where the 12 month moving average stands at 5.2%. If the unemployment rate increases by 0.1% each of the next two months (April and May – remember the reported unemployment rate is for the previous month) then the rate would cross above the 12 month moving average. We won’t find out until the May unemployment rate  is reported at the beginning of June. And we’ll know this week what the April rate is. This seems unlikely but you never know. The FOMC’s own projections don’t support a change but they are notoriously poor forecasters. Others think that more realistically the end of the year would be the time frame we could possibly see a trigger. But there is no need to forecast to use the UI system. For now, if you were using this system it would be risk on still.

In summary, historically the change of trend in the unemployment rate has been a good signal to time the market. Better than the two most popular trend indicators around.

 

 


18 Comments

Graham · May 3, 2016 at 8:40 am

did you lag the data to account for forward looking bias when testing?

    paul.novell@gmail.com · May 4, 2016 at 8:53 am

    No need. Data is already lagged when reported. See the source post for more. Also ran the sims on every day of the month – same result.

    P

John · May 3, 2016 at 8:42 am

Paul,

Thank you for bringing this to our attention. It looks like a very interesting approach.

My reservations are that the reported unemployment rate is debatable as to whether it reflects actual unemployment data vs massaged data for general consumption.

Data accuracy is slowly but surely becoming an important issue for market evaluation.

    paul.novell@gmail.com · May 4, 2016 at 8:51 am

    Maybe. That’s not what the data says.

Govind · May 3, 2016 at 9:41 am

Paul – Since SPY began trading in 1993, why did you wait until 1999 to do the comparison?

    paul.novell@gmail.com · May 4, 2016 at 8:49 am

    The software I use only has data back to 1999.

    Paul

Tony · May 3, 2016 at 10:03 am

Wow! This is truly awesome stuff Paul! Really looking forward to your follow-up posts on this.

Specifically, I would be interested in any backtesting you have looked at with the UI and the various quant strategies. When I have backtested risk-off valuation indicators like CAPE ratios they look good on the SPY, but they don’t work well with the quant systems.

The quant systems tend to have lower drawdowns during recessions than an index, and also massive outperformance when things are good. The traditional indicators I have looked at have you sitting on the sidelines during a lot of that outperformance, and only save you occasionally from the big downturns. Better to just always be invested in the quant systems regardless of the environment. Or so I thought…

    paul.novell@gmail.com · May 4, 2016 at 8:49 am

    You’re right. Traditional indicators like the 200 day SMA on the SP500 don’t work well on quant strategies. In light of that bette to always be invested in quant strategies or use other means such as trailing stops to reduce drawdowns. We’ll see how this indicator works on quant strategies.

Andrew · May 3, 2016 at 11:52 am

Very interesting! I would have assumed the UI was more open to “fudging” in reporting, and so less reliable, whereas the price signal used in SMA can’t.

    paul.novell@gmail.com · May 4, 2016 at 8:47 am

    Maybe. That’s not what the data says.

    P

Don · May 3, 2016 at 12:35 pm

Fantastic info. Thanks.

Lars · May 3, 2016 at 3:41 pm

Hi Paul,

Awesome stuff, and very timely, since I’ve also just read that article by “Philosophical Economics” last week, and have since been thinking about how to possibly apply that to the GTAA portfolios. Seems pretty straightforward for the GTAA13 portfolio (apply to the domestic equity asset signals only?), but a bit less obvious for the AGG3/6 portfolios because of their asset rotation rules. Looks like this additional filter could have potentially prevented the most recent whipsaws last August and this January, which each cost these portfolios a few percentage points of performance.

Looking forward to your follow-up posts!

-Lars

    paul.novell@gmail.com · May 4, 2016 at 8:46 am

    That would be one way to do it, yes.

    Paul

Richard Wilkes · May 3, 2016 at 8:06 pm

Paul, great observation and correlations! Indeed having other metrics to overlay on the in or out scenarios is very compelling. Thanks for your contributions!

Wayne · May 4, 2016 at 12:54 am

Thanks for posting! It was interesting and I will be paying attention in the future. Where did you find the unemployment rate information? I would be interested in doing some of my own studies.

    paul.novell@gmail.com · May 4, 2016 at 8:45 am

    Thanks Wayne. Best place is to use FRED
    Or the BLS website.

    Paul

Niklas · May 6, 2016 at 1:05 pm

Hi Paul,

thanks for the good post! It was interesting to read. I also use some forward looking parameters for my strategies. If I have understood correctly, you use porfolio123 for backtesting. It would be interesting to know what is the slippage you use for testing the strategies? My experience is that the slippage has massive impact for any backtesting results. At least GTA13, AGG6 and AGG3 (and all active strategies) show completely different performance if you use i.e. 1% slippage vs. no slippage at all. I have to say that I’m not at all sure if these strategies are so good as it seems with the first glance if you take the slippage into account. What do you think?

Jeff · May 7, 2016 at 8:52 am

Very exciting Paul, I can definitely see myself adopting this strategy and greatly look forward to your future posts. Thank you for all of your efforts!

Comments are closed.