Market Timing: A New Systemic, Multi-Valued Indicator (Part 3)

| About: SPDR S&P (SPY)

Summary

MTS10 is a risk indicator taking into account technicals, fundamentals, sentiment and economy.

It may be used for scaled hedging or market timing.

It has done a good job as a timer in the last 17 months.

Part 2 was published here on 5/21/2015, 17 months ago. This article reports how my systemic risk indicator has worked since then.

Since I have not written about this for a long time (except to subscribers who receive an update every week), a reminder of the philosophy and context is necessary.

There is no single safe market-timing indicator.

Imagine you find an indicator predicting all eight recessions in the United States since 1950 without giving a wrong positive on the whole period. Is it the holy grail of market timing? Not really. In statistics, being right eight times out of eight implies being right 67% of the time with a 95% confidence interval. In other words, there is a 95% probability that the indicator's accuracy is at least 67%. How much of your portfolio would you bet on an indicator that may be wrong 33% of the time?

Data to test the usual market timing indicators are available for about a century, in the best cases. The United States has crossed 22 recessions since 1900. This data sample is too small to claim that one timing indicator is safe enough to make a large bet.

From Market Timing to Systemic Risk

I have designed a Market-Timing Score using 10 indicators: MTS10. It counts the bearish signals given by these indicators, so it takes an integer value between 0 and 10. The components are based on various data series: S&P 500 price (death cross), market breadth (% of stocks technically bearish), average short interest of S&P 500 companies (two trends based on different look-back periods), U.S. unemployment rate (two trends), S&P 500's current-year EPS estimate (two trends), S&P 500's trailing 12-month EPS, and new house starts.

The choice of these indicators is not random. It is backed by research, and covers four fields of market analysis: technicals, sentiment, fundamentals, economy. Using various data series and trends is not for optimization purposes, it is even the opposite. Indeed, some economic data are sometimes revised, and there is no public database with historical point-in-time data. Workarounds are:

  • Including more indicators to lower the influence of a false signal.
  • Using moving averages to lower the impact of a revision.
  • Increasing the number of possibly flawed indicators by setting an alarm level (explained below). Risk management is not hard science. A flexible systemic model coping with uncertainty is safer than perfect pseudo science.

The next table shows the annualized return, standard deviation of weekly return and maximum drawdown of SPY from 1/1/2001 to 10/23/2016, going in cash when MTS10 is above a value 'n' (using the current economic data, with the look-ahead bias of revisions).

Portfolio in SPY when MTS10<=n, in cash above n

n= 0

1

2

3

4

5

6

7

8

9

10

Annual. return %

2.13

3.37

6.10

6.64

7.61

10.88

11.13

9.43

7.28

7.97

5.25

Stand. deviation %

4.51

7.37

8.78

9.91

10.91

11.81

12.49

12.84

13.33

14.99

18.05

Max. Drawdown %

13.50

14.98

19.22

20.15

20.15

20.15

20.15

31.18

40.33

44.02

55.42

Click to enlarge

The return goes up until 6, then down with a significant increase in risk measured by standard deviation and drawdown.

Out-of-sample results

When publishing this table in May 2015, I proposed two ways to use MTS10: as a multi-valued indicator in a scaled hedging strategy or as a binary indicator by going out of the market or taking a full hedge above a specified alarm level. I suggested MTS10>=7 as an alarm level. Past data are no guarantee against false signals in the future (MTS10>=7 and the market goes up) or black swans (MTS10<7 and something bad happens). MTS10 is like the avalanche danger scale for skiers and snowboarders. It measures objectively how dangerous an environment is to decide about taking risks. It is not a guarantee that nothing bad can happen when the danger is low.

The next table shows how MTS10 alarm level "7" has worked as a market-timing indicator between 5/23/2015 and 10/8/2016 on the S&P 500 index return (without dividends). It compares it with two popular bearish signals on the S&P 500 index: the death cross (50-day sma below the 200-day sma) and the spot price below the 200-day sma.

going Cash when...

never

MTS10>=7

S&P500 death cross

S&P500 below 200dsma

S&P 500 return

1.69%

4.37%

-10.3%

-7.07%

Click to enlarge

On this out-of-sample period, MTS10>=7 has done much better than the usual technical timers, which have been whipsawed. It has even brought a small excess return. There is no guarantee on future results, but it's good to see that it has done a decent job since its introduction in all my weekly services. As noted above, some components are subject to revisions. The return is calculated from the real MTS10 values sent every week to subscribers, not from the readings currently in the database, which have a look-ahead bias.

The simplified MTS4 indicator was less accurate. With an optimistic alarm level (MTS4=4, never hit on the period), the return is equal to the index without timing at 1.69%. With a pessimistic alarm level (MTS4>=3), the return is -8.8%, ranked between the two usual technical timers.

Once again past performance is not an indication of future results. My focus is less about backtests and real past performance than about a way to aggregate multiple views of risk to cope with complexity and uncertainty in the future.

Disclosure: I am/we are long SPY.

I wrote this article myself, and it expresses my own opinions. I am not receiving compensation for it (other than from Seeking Alpha). I have no business relationship with any company whose stock is mentioned in this article.

Additional disclosure: I am long SPY in an ETF model based on MTS10 but net short S&P 500 for hedging purposes.