Dissecting Goldman’s 99 percentile Market-Timing Signal

/Dissecting Goldman’s 99 percentile Market-Timing Signal

Dissecting Goldman’s 99 percentile Market-Timing Signal

Print Friendly, PDF & Email
(Last Updated On: August 18, 2017)

Investors have been worrying, at least for the last several years, that the market is overvalued. By some measures this is undoubtedly true. Just yesterday we highlighted that the Shiller CAPE is in the 94th percentile as of 1/31/2015.

And as valuations have gone higher, the alarm bells in the press have gotten louder.

Digging into the Goldman Sachs call is illuminating. Here is what Goldman’s chief strategist David Kostin says on current market valuations:

Stocks with attractive valuation are rare in the current environment of stretched share prices. The aggregate S&P 500 trades at 17.3x forward EPS and 10.2x EV/EBITDA. The only time during the past 40 years that the index traded at a higher multiple was during the 1997-2000 Tech Bubble. The median stock sports a P/E and EV/EBITDA of 18.0x and 11.0x, respectively. These valuations rank in the 99th percentile of both P/E and EV/EBITDA multiples since 1976.

The implicit assumption underlying Kostin’s “story” is that knowing that the market is in the 99th percentile somehow improves our ability to time the market.

A great story–but is there any evidence to support this claim?

Does Goldman’s 99% Valuation-Timing Rule Work?

We’ve analysed tactical asset allocation using valuation measures in the past.

The evidence isn’t promising: Trend-following timing rules have been much more effective than valuation-based timing rules.

Nonetheless, the Goldman Sachs article inspired us to dig into the valuation-timing hypothesis.

To create our “valuation-timing” indicator, every month we identify the 99 percentile valuations using rolling 5-, 10-, and 20-year look-back periods. Our trading rule is simple: if the current market valuation is greater or equal to the 99 percentile measure, we invest in the risk-free rate (short-term treasury bills), otherwise, we stay invested.

We compare the valuation-timing indicator to a monthly-assessed simple moving-average (MA) trading rule, and a buy-and-hold strategy. The buy-and-hold strategy is straightforward, and the MA indicator is simple: if the current market price is lower than the 12 month moving average, we invest in the risk-free rate (short-term treasury bills), otherwise, we stay invested.

Our conclusion is counterintuitive, but not entirely surprising:

Goldman’s “Valuation-Timing” concept doesn’t have legs.

Strategy Details:

For the MA signals, we use a monthly-assessed 12-month MA rule on the S&P 500 total return index.

  • If the price for last month is above the past 12 months average, stay in the market; otherwise, invest in the risk-free asset.

For the valuation signal, we use CAPE, Shiller’s Cyclically Adjusted PE ratio. Results are similar for P/E and enterprise multiples, but for public replication purposes we use the CAPE data (we sometimes make mistakes and want others to let us know!). CAPE raw data can be accessed from Shiller’s database.

  • If last month’s CAPE valuation is in the 99 percentile or higher, buy U.S. Treasury bills (Rf), otherwise stay in the market. For robustness purposes, we use three different rolling look-back periods to determine the 99 percentile valuation at a given point in time: 5-, 10- and 20- years.

Our backtest period is from 1/1/1947 to 1/31/2015 (we start in 1947 because we need to burn 20 years of data for the 20-year look-back metric). Results are gross, no fees are included, and only index returns are included. Indexes are unmanaged, do not reflect management or trading fees, and one cannot invest directly in an index. All returns are total returns and include the reinvestment of distributions (e.g., dividends). Strategies are all monthly-rebalanced, meaning Goldman’s valuation litmus test is applied every month.

Strategy Legend:
  • SP500 = S&P 500 Total Return Index
  • LTR = The Merrill Lynch 10-year U.S. Treasury Futures Total Return Index
  • Rolling 5 year 99perc CAPE= Timing signal uses the 99th percentile valuation metric using rolling 5 year look-back periods.
  • Rolling 10 year 99perc CAPE = Timing signal uses the 99th percentile valuation metric using rolling 10 year look-back periods.
  • Rolling 20 year 99perc CAPE= Timing signal uses the 99th percentile valuation metric using rolling 20 year look-back periods.
  • (1,12) MA= If last month’s price is above the past 12 month average, invest in the S&P 500; otherwise, buy U.S. Treasury Bills (RF).

Statistics Summary:

Full Sample: 1/1/1947 – 1/31/2015
  • (1,12) MA outperforms Rolling 10 year Valuation Timing.
  • Buy-and-hold is similar to Valuation-Timing.
1

The results are hypothetical results and are NOT an indicator of future results and do NOT represent returns that any investor actually attained. Indexes are unmanaged, do not reflect management or trading fees, and one cannot invest directly in an index. Additional information regarding the construction of these results is available upon request.

First Half: 1/1/1947 – 12/31/1981
  • (1,12) MA outperforms Rolling 10 year Valuation Timing.
  • Buy-and-hold is similar to Valuation-Timing.
2

The results are hypothetical results and are NOT an indicator of future results and do NOT represent returns that any investor actually attained. Indexes are unmanaged, do not reflect management or trading fees, and one cannot invest directly in an index. Additional information regarding the construction of these results is available upon request.

Second Half: 1/1/1982 – 1/31/2015

  • (1,12) MA outperforms Rolling 10 year Valuation Timing.
  • Buy-and-hold is similar to Valuation-Timing.
3

The results are hypothetical results and are NOT an indicator of future results and do NOT represent returns that any investor actually attained. Indexes are unmanaged, do not reflect management or trading fees, and one cannot invest directly in an index. Additional information regarding the construction of these results is available upon request.

Robustness Tests with Different Look-Back Periods

Full Sample: 1/1/1947 – 1/31/2015
  • (1,12) MA outperforms Rolling 5-, 10-, and 20-year Valuation Timing.
  • Buy-and-hold is similar to Valuation-Timing.
5

The results are hypothetical results and are NOT an indicator of future results and do NOT represent returns that any investor actually attained. Indexes are unmanaged, do not reflect management or trading fees, and one cannot invest directly in an index. Additional information regarding the construction of these results is available upon request.

First Half: 1/1/1947 – 12/31/1981
  • (1,12) MA outperforms Rolling 5-, 10-, and 20-year Valuation Timing.
  • Buy-and-hold is similar to Valuation-Timing.
6

The results are hypothetical results and are NOT an indicator of future results and do NOT represent returns that any investor actually attained. Indexes are unmanaged, do not reflect management or trading fees, and one cannot invest directly in an index. Additional information regarding the construction of these results is available upon request.

Second Half: 1/1/1982 – 1/31/2015
  • (1,12) MA outperforms Rolling 5-, 10-, and 20-year Valuation Timing.
  • Buy-and-hold is similar to Valuation-Timing.
7

The results are hypothetical results and are NOT an indicator of future results and do NOT represent returns that any investor actually attained. Indexes are unmanaged, do not reflect management or trading fees, and one cannot invest directly in an index. Additional information regarding the construction of these results is available upon request.

Robustness Tests with 75 percentile Buy-In Trigger

One issue with a 99th percentile trigger is the possibility that you get out when valuations are at 99th percentile, but when the market drops a bit and hits 98th percentile you get right back in. This might “whipsaw” and create poor outcomes.

To test this conjecture we do another test where we apply the 99th percentile rule, but we don’t get back in to the market until it has dropped to at least the 75th percentile.

FULL SAMPLE: 1/1/1947 – 1/31/2015
  • (1,12) MA outperforms Rolling Valuation Timing with a 75th percentile buy-in rule.
  • Buy-and-hold is similar to Valuation-Timing.
9

The results are hypothetical results and are NOT an indicator of future results and do NOT represent returns that any investor actually attained. Indexes are unmanaged, do not reflect management or trading fees, and one cannot invest directly in an index. Additional information regarding the construction of these results is available upon request.

FIRST HALF: 1/1/1947 – 12/31/1981
  • (1,12) MA outperforms Rolling Valuation Timing with a 75th percentile buy-in rule.
  • Buy-and-hold is similar to Valuation-Timing.
10

The results are hypothetical results and are NOT an indicator of future results and do NOT represent returns that any investor actually attained. Indexes are unmanaged, do not reflect management or trading fees, and one cannot invest directly in an index. Additional information regarding the construction of these results is available upon request.

SECOND HALF: 1/1/1982 – 1/31/2015
  • (1,12) MA outperforms Rolling Valuation Timing with a 75th percentile buy-in rule.
  • Buy-and-hold is similar to Valuation-Timing.
11

The results are hypothetical results and are NOT an indicator of future results and do NOT represent returns that any investor actually attained. Indexes are unmanaged, do not reflect management or trading fees, and one cannot invest directly in an index. Additional information regarding the construction of these results is available upon request.

Conclusion

There is no evidence to support the use of “valuation-timing,” which performs similarly to buy-and-hold strategies (after costs it would we much worse). There is nothing magical about the 99th percentile. Trend-following, at least historically, seems to more effective.

Perhaps there are more convoluted, complex, and data-optimized ways in which we can leverage overall market valuations to help us time markets. We haven’t found any, but that doesn’t mean they don’t exist. Please share.


  • The views and opinions expressed herein are those of the author and do not necessarily reflect the views of Alpha Architect, its affiliates or its employees. Our full disclosures are available here. Definitions of common statistics used in our analysis are available here (towards the bottom).
  • Join thousands of other readers and subscribe to our blog.
  • This site provides NO information on our value ETFs or our momentum ETFs. Please refer to this site.

About the Author:

After serving as a Captain in the United States Marine Corps, Dr. Gray earned a PhD, and worked as a finance professor at Drexel University. Dr. Gray’s interest in bridging the research gap between academia and industry led him to found Alpha Architect, an asset management that delivers affordable active exposures for tax-sensitive investors. Dr. Gray has published four books and a number of academic articles. Wes is a regular contributor to multiple industry outlets, to include the following: Wall Street Journal, Forbes, ETF.com, and the CFA Institute. Dr. Gray earned an MBA and a PhD in finance from the University of Chicago and graduated magna cum laude with a BS from The Wharton School of the University of Pennsylvania.
  • Good work. More proof showing that valuation levels derived from “fundamental” metrics can confirm, but not be included towards, a sound, non “fundamentals” based tactical asset allocation model’s buy and sell decisions.
    An improvement that I have brought to light regarding an SP 500 moving average rule methodology is the addition of bond allocation during “T-bill” allocation. When the SP 500 is below it’s 10 mo SMA * or in “T bills” ( M. Faber’s parameter ) then invest in intermediate / long bonds if / when the price of bond > than it’s 10 mo SMA … Sell when bond moves than 10 mo SMA …

    * M. Faber’s parameter

  • Roger James

    I think this demonstrates very clearly that we need to be sceptical of anything that Goldman Sachs says. They only exist to make money for themselves and incidentally for their clients. There will ALWAYS be an ulterior motive behind their pronouncements. Perhaps they are short the market at the moment?

  • Corpraider

    Hi Wes,

    I really enjoy your blog and other tools on the site and am a shareholder in QVAL. I wonder if you’ve looked at perhaps using a high market valuation combined with the momentum tactical overlay. For example, applying the 260 day SMA rule, but only when the CAPE is above whatever deviation from the norm you think could serve as a reasonable proxy for highly valued (maybe the 75% expensive could work)?

  • Few ideas we have explored that are somewhat related:

    http://www.alphaarchitect.com/blog/2014/06/12/can-market-valuations-be-effective-market-timing-signals/#.VPZD1vnF9MU

    http://www.alphaarchitect.com/blog/2014/08/22/tactical-asset-allocation-during-cheap-markets/#.VPZD0_nF9MU

    We’re working on some more integration research at the moment. We’ll post what we find.

  • Corpraider

    Thanks! Yeah seems like you’ve looked at some pretty similar stuff. I read an AQR paper on a combo but I think they just modified allocations. Seems like, if you view the SMA as a risk management tool, there might be some logic to applying it when values are elevated, but I guess that won’t make the data spring into existence. hah.

  • puneet singhania

    Have u taken 1 year forward earnings or ebdita….if not then the backtest done is not based on how goldman uses this metric and the backtest is not correct.

  • Puneet, thanks for the question.
    We didn’t do fwd earnings, but we’ve done all the TTM across a variety of valuation metrics — same results.
    But I don’t think that what we highlight in this quick article is a heroic finding — there are few researchers who find evidence that market-wide valuation metrics are highly effective for tactical short-run timing.

  • Brett House

    I don’t see why you adjusting the CAPE or 1/CAPE with inflation. In the downloadable file that Shiller provides online it’s clear that the price series is in real terms and the earnings ratio is also in real terms.

  • Agree, but where did we mention an inflation adjustment in this study? Are you referring to a different study?

  • Brett House

    Sorry, I may have queried in the wrong thread. I noticed the inflation adjustment in your related post here: https://alphaarchitect.com//2015/07/21/eureka-a-valuation-based-asset-allocation-strategy-that-might-work/#gs.BoEH038

  • Hey Brett,
    The 1/CAPE ratio represents a “real yield” to investing in equity — as you point out. The measure then takes off a realized inflation rate. The idea is this is the “real yield” after taking out realized inflation, leading to a “real equity premium.” When this is high, this is hypothesized to be good for equity, and vice versa.
    That is the story the Gestaultu folks tell, at least. I think it is reasonable. My concern is not with the story, but with the fact that there are so many “valuation timers” that don’t work at all. So by random luck one could generate a valuation timer that works well. Hard to handicap if this valuation timer works because it is robust or because it was a lucky choice…we’ll never know I guess.

  • Brett House

    Thanks for your reply. What I specifically don’t understand is why the Gestaultu folks are subtracting a realized inflation rate from 1/CAPE when 1/CAPE is already composed of a real earnings measure and a real price measure. What is the rationale? It makes sense to subtract, say, a 10yr UST yield from 1/CAPE to get the equity risk premium. But I have no idea what 1/CAPE – inflation is meant to be. 1/CAPE is already the real equity yield; there is no need or economic rationale to subtract inflation from it. Can you provide any additional insight? I wrote to Gestaultu but it does not reply.

  • Agree, somewhat akin to looking at price-to-earnings and enteprisevalue-to-earnings. P/E makes “sense” where EV/E doesn’t. But their measure works well…so now you are in the game of determining whether the signal is driven by a robust underlying process or is it luck/noise.