Updates Posted for Week Ended May 17, 2013

Another week, another new high for US equities… the week saw most asset classes drift in the direction of their 13-week momentum. US equities were up over 2%, global real estate over 3%, and most other asset classes meandered in the general direction of their recent trend. Commodities and emerging markets were down slightly, Canadian equities and real estate were barely positive, and high yield bonds gave up a bit of ground. 

Not much changed in the models’ positioning. The recent strength of global real estate, as well as an easing of interest rates, bumped them up to an overweight in both Canadian and US models – in the former case, taking capital away from Canadian real estate, in the latter, from cash.

It’s a long weekend here in Canada, so I will try to find time to post some additional analysis, but in case other priorities intervene, I’ll sign off with, “Have a good week!”

Surprise! Weekly Updates Posted for Week Ended May 10, 2013

John Hussman’s weekly market commentary for April 29, “When Rich Valuations Meet Poor Economic Data,”  includes a new look  at an interesting indicator, the Citi Economic Surprise Index. This indicator tracks the degree to which announced economic data exceed or miss economists’ forecasts; the indicator value is a running total of the cumulative positive and negative surprises, weighted according to their importance, based on their historic impact on markets. The index has been calculated for the US and various other geographies back to 2003.

Hussman Funds’ Bill Hester analyzed the history  of the indicator’s correlation with equities (specifically using the US version of the index, CESIUSD, and the S&P 500 index). Hester’s analysis found  that periods of  strongly negative correlation seemed to anticipate abrupt market declines, particularly once the negative correlation started to reverse:

Figure 1.  CESIUSD Correlation to the 13-Week Change in the S&P 500 Index, and S&P 500 Index

Data sources: Bloomberg, Yahoo! Finance

The indicator is not perfect – witness the rise in markets following the May-June, 2012 signal – but the illustration suggests that when the US stock market runs up in the face of weakening economic data, as it has done in recent months, this tends to end badly… If we summarize the results graphically, we can see  the following:

Figure 2. SPX 4-Week Return For Extremes of Correlation, and Direction of CESIUSD

Data sources: Bloomberg, Yahoo! Finance

This looks equally compelling, but if we split out the sample set, the results look much less unequivocal. Although the example shown above (and the one used in the article on the Hussman Funds page) focuses on post-2008 data, the effect was actually stronger pre-2008:

Figure 3. Forward Returns by Level of Correlation, Pre- and Post-2008

Data sources: Bloomberg, Yahoo! Finance

if we look beyond the borders of the US, we can see that even for the full history of the indicator, the relationship between correlation and subsequent returns is completely different.

Figure 4. Forward Returns by Level of Correlation, Developed and Emerging Markets, 2003-2013

Data sources: Bloomberg, Yahoo! Finance

What looked like a promising indicator (in the sense of having strong predictive power, not in terms of foreshadowing positive returns!) turns out either to work only under certain conditions (ie, in the US, and particularly before 2008). What could be going on here?

My guess (and it is only that) would be that we have an example here of a set of indicators that were derived from mining a particular data set (US data, 2003-2007 or so), and that do not generalize very well to other markets or time periods. Although the CESI indices have been calculated back to 2003, I don’t believe that they were published that far back – I believe (please correct me if I’m wrong) that they first appeared around 2007. This could mean that they were calibrated using a sample of 2003-2007 US data, with the 2008-13 being the live, “out of sample” data. As we have seen, the results post-2008 are considerably weaker, and the indices do not apply very well outside of the US, which supports the hypothesis that the indices are a bit US-specific, and perhaps the result of expecting history to repeat itself a little too closely.

From a general Market Compass perspective, this is the kind of finicky, only-works-sometimes indicator that we would prefer not to rely on in ever-changing (but always rhyming) financial markets.

The weekly updates have been posted. The models made a few small changes. Weakness in the momentum measure for Canadian equities resulted in a decreased weighting, which went to Canadian real estate securities, whose modest pullback has improved their value score. The US model saw small caps take back a bit of capital from international stocks as they traded places in the momentum rankings. Generally, though, the models remain long everything risky, with the strong exception of commodities, where they remain at zero weight.

Model outputs are available at the usual place.

Have a good week!

O Canada! Updates for Week Ended May 3rd, 2013

New highs on the S&P 500! A surprisingly okay US employment report! Nearly 50% of S&P 500 companies beating earnings expectations! The headlines from the US have been consistently upbeat over the past week, and as the weekend rolls around, the weekly models deliver their verdict… increase allocation to Canada!

No, this is not the result of some profound macroeconomic thesis (strong US economy means demand for Canadian goods… buy Canadian stocks!) but rather a simple matter of trend following. Canadian stocks had been showing weakening trend measures (not helped by the, er, melt-down of the gold industry), but have recovered in recent weeks. Canadian stocks return to a full score on trend, and that, coupled with a bonus for cheap valuation, put them at an overweight.

All this over- and underweighting concerns only the weekly models (and only the Canadian model, at that). The monthly models have also been updated, and there, the story is much less dramatic. There is no change to the monthly models’ positions; they remain fully invested across all regions. The monthly model is meant to be a low-activity model for people who don’t want to make many changes to their accounts, and so far it has performed exactly as intended.

The model outputs can be found as always below the “Current Model Outputs” tab.

Have a good week!

When Bad News is Good News: Updates for Week Ended April 26, 2013

This week saw the release of several disappointing economic numbers. Most prominently, the preliminary estimate of US GDP growth for Q1 came in at 2.5%, missing the median  forecast of 3.0%. GDP is a difficult indicator to use in real-time decision-making, though, as it is released with a significant lag (the preliminary release is nearly a month after quarter-end) and it is then revised heavily – occasionally by more than a full percentage point. So whether a miss of 0.5% is meaningful, or even stands in the final tally, remains to be seen.

A more timely barometer of economic conditions is provided by the various purchasing managers’ surveys that attempt to measure business conditions for the current month by asking businesses about current and expected levels of activity. The survey results are summarized as a number centered around a threshold of 50, where readings above 50 indicate expansion of activity and vice versa. This week saw the release of April PMI results for a number of countries and regions, including China, Europe and the US (the US actually offers several different PMI surveys, of which one was released this week). Although there were a couple of small positives, the overall the picture was poor. Reported numbers missed expectations in a majority of markets, regions that are in economic expansions (China and the US) both showed signs of slowing, and although there were some marginal spots of improvement in Europe, not a single economy showed signs of pulling out of contraction.

In short, the data were consistent with the story of recent quarters: slowing growth in China, weak recovery in the US, and recession in Europe. There were some positives – US weekly jobless claims were down, and housing sales continued to suggest a creeping recover – but overall there were more misses than hits for the week.

On top of all this, US corporations continued to release results for the first quarter, and these were consistent with the trend in recent years: a reasonable proportion (roughly 60%) beating earnings expectations, but more than 50% missing revenue targets. Low growth, cost-cutting and some prominent warnings about slowing demand from China combined to paint a somewhat cloudy picture for the outlook for US corporations.

Against this gloomy backdrop, risk markets rallied around the world:

Figure 1. Returns for Selected Markets

Data Source: Yahoo! FInance

Almost every asset class delivered positive returns for the week; even commodities and Canadian small-cap stocks came in on the positive side. A minor give-back from global real estate seems hardly meaningful after the strong returns that this asset class has delivered over the last 13 weeks.

What could be going on here – the news is bad, but markets around the world rally?

The conventional explanation is that bad economic news is good for markets in that it ensures that central banks will maintain their highly accommodative policies for longer, keeping the monetary taps open and providing further fuel for appreciation of risk assets. This is possible, although difficult to demonstrate conclusively.

Market Compass’ approach to weeks like this is… not to try too hard to explain every counterintuitive gyration of the markets and economies. We register the weak data in the models and move forward, recognizing that markets will often behave in ways that don’t appear to make sense.

This is part of the reason why the models employ a trend-following component: if markets are going up, we want the models to be invested, even if this seems to be against all reason. The models will reduce their allocation to a market if the macro situation gets particularly weak (as is the case right now in Europe) or if valuations get very stretched (which we see in US stocks or high yield bonds) but they will retain some exposure as long as the trend is up.

We saw some trend-driven changes in the models’ positioning this week. The Canadian model reduced its exposure to Canadian stocks as momentum turned very weak, compounding the weakening trend of previous weeks. Emerging markets, on the other hand, returned to full strength on our trend measures; this, combined with very favourable valuations, moves them to an overweight in both the Canadian and US models.

So it was not only the markets at large that took economic bad news and interpreted it as good: the Market Compass models also responded to the week by increasing their allocation to risky assets by 5%.

The model outputs can be found as always under the “Current Model Outputs” tab.

Have a good week!

Updates Posted for Week Ending April 19, 2013

The weekly updates have been posted. Both Canadian and US models have reduced market exposure by a small amount for the week. In the case of the US, the adjustment was 3%, driven by a downgrade of the US’ momentum score as the remarkable run of the first quarter has slowed. This adjustment is not that surprising – momentum comes and goes (which is why it falls under the “tacking and gybing” heading in our scorecard). The slower-moving “steady as she goes” indicators of macro and trend are responsible for a greater proportion of the models’ positioning and results over time, and it is on the basis of one of these that the Canadian model reduced exposure. For the first time in some months, the strength of the uptrend in Canadian equities has weakened, leading the Canadian model to reduce exposure by 5%. We can see the weakness of the trend reflected in the chart below:

Figure 1: Price Returns for Key Asset Classes

Data source: Yahoo! Finance

As can be seen, Canadian equities show among the weakest return over most time periods shown. The kicker this week for Canadian equities was that the average of 3-, 6- and 12-month price change slipped below zero, leading the models to reduce weighting based on the trend factor.

Looking more broadly at the asset classes shown above, the worst performers overall have been Commodities (which of course are a major driver of Canadian equity returns), and Canadian small caps, which often serve as a good barometer for investor enthusiasm for resource stocks. Emerging Markets have also been weak. All of these asset classes have for the past decade served as a good indication of investor confidence in global economic growth. At this point, the trends captured by the models suggest that confidence is waning, with commodities at zero weight and both Canadian and Emerging Markets stocks underweight on trend weakness.

The models highlight several other areas of concern – International equities face macroeconomic headwinds (in Europe in particular) and valuations look stretched in US small caps, High Yield and some areas of real estate. These are summarized on the model output scorecards, which can be found under the “Current Model Outputs” tab above.

As market exposure has come down (from nearly 100% in December to roughly 75% now), the question arises of what to do with the cash that this has released. For Canadian investors, one reader recommended the attached article from Rob Carrick, which identifies a number of options for a world in which broker interest and retail money market yields are basically zero:

http://www.theglobeandmail.com/globe-investor/investment-ideas/when-cash-is-no-longer-king/article4896342/

For US readers, the Fed has made the situation even more difficult. If any readers have suggestions on substitutes for traditional cash investments, please feel free to share.

Have a good week!

All that Glitters

The past week included a number of unusual events. North Korea continued its sabre rattling. The US Federal Reserve released its closely-watched meeting minutes nearly a full day early to a variety of Wall Street banks, providing a field day for conspiracy theorists, before “correcting” the error the following morning by releasing the notes more widely. And gold suffered one of its worst weekly declines ever, falling over 5% in US dollar terms, putting it, as many commentators noted, in “bear market” territory as it has now fallen over 20% from its peak of over $1900 per ounce.

The asset class summary for the week shows the impact of this fall (and that of the price of oil) on commodities, which fell nearly 4% for the week. Canadian and Emerging Markets stocks were likewise dragged down, whereas US and developed market stocks delivered small gains, helped, perhaps, by the dovish Fed minutes.

Figure 1. Weekly Returns for Selected Asset Classes

Data source: Yahoo! Finance

Market Compass does not allocate directly to gold or gold stocks, although there is some exposure to each embedded in the broader indices that the models track. There are several reasons for this: historically, gold and gold stocks have historically behaved quite differently from broader aggregates of commodities and stocks, probably at least in part because of the supposed “safe haven” qualities of precious metals and related assets. In practice, gold and gold stocks do not deliver particularly strong results under trend-following approaches like the one that is at the core of Market Compass, so including it in the models does not offer much benefit. Holding gold as “disaster insurance” or as an inflation hedge may be psychologically comforting, but the empirical evidence for its effectiveness in these roles is mixed at best.

However, the recent sell-off was pronounced enough that Market Compass dug into the question of whether there might be some mean reversion potential in gold or gold equities. In other words, when they sell off this much, do the metal or the stocks offer some potential to bounce back?

To put last week’s declines into perspective, here is the price action of GLD, the most popular gold bullion ETF, and GDX, the MarketVectors Gold Miners ETF:

Figure 2. GLD Price and 13-Week Rate of Change

The blue line shows the ETF’s price, and the red columns the trailing quarterly (13-week) change in percentage terms. As we can see, the ETF has fallen as rapidly as 10% per quarter in recent weeks, a rate that it has exceeded on two other occasions since it began its retreat from over $180 per share. The only time when its decline was more abrupt was during the crisis period in 2008-9 (again suggesting limited effectiveness of holding gold in the face of distressed markets).

Figure 3. GDX Price and 13-Week Rate of Change

GDX, the gold miners ETF, shows even more striking declines, backing up the view of those who see gold stocks as a leverage play on the gold price. The correlation is far from perfect, however – gold equities lagged the metal in its climb to $1900, and have underperformed it in its decline. The amount of red under the 0% line is considerably greater in this chart than in GLD’s.

Looking at the pattern of what happens after declines of these magnitudes, we took a simple look at whether there is typically a reliable bounce from declines of this depth. This decline in both gold and gold stocks has been extreme – the move in both represented a decline of 2 standard deviations of weekly returns. Since the inception of the GLD ETF, there have been only 9 other occasions where both metal and miners have declined by 2 standard deviations, all in 2008. The return distribution for the following 4 weeks looks like the attached:

Figure 4. 4-Week Returns Following 2 Standard Deviation Declines in both GLD and GDX

 

History – based on very few occurrences – suggests that there may be an interesting gamble here for the brave. Roughly half of the occurrences lead to a positive outcome, with the average gain significantly outweighing the average decline. Looking out over 13 weeks, the results appear even more skewed to the upside:

Figure 5. 13-Week Returns Following 2 Standard Deviation Declines in both GLD and GDX

It is important to note, however, that this illustration not only represents a very small sample, but also shows results from only a single year – and ultimately, from a single crisis. The time periods in the above chart overlap, so these are not statistically independent events. So a reasonable way to interpret these charts is to say that “in 2008, gold and gold stocks got hammered, and then bounced.” Whether the same thing will happen again is impossible to conclude definitively on the basis of one violent decline.

More fundamentally, aggressive dip buying is not part of the Market Compass approach, so the models will not take a trade on the basis of this kind of analysis. Should a market decline like the one in precious metals reverse itself into an uptrend, there could be a trade for a Market Compass-type approach – ie, once trend-following and mean-reversion line up. But for now, the models remain focused on asset classes that tend to behave a bit better on average, and most of which are, for now at least, in reasonably solid uptrends.

The models made a few changes this week. The US model reduced its stock market exposure slightly up as US large caps lost a bit of momentum. The Canadian model saw a shift of several percent from US equities to International, reflecting the continued surge of Japanese markets. Both models hold significant levels of cash (20% in the Canadian model, 25% in the US version), with the largest contributor still being the 0 weight to commodities, which certainly seemed the right positioning this week!

The model outputs are available as always at:

http://marketcompass.wordpress.com/current-model-outputs/

Have a good week!

US or World?

A reader wrote recently asking an interesting question prompted by some of the practical issues associated with ETF investments. The question was (in my paraphrase): for an investor who wants to hold only one investment, (for example, in a small tax-sheltered account such as a Canadian TFSA or a US 401 (k)), given the lower cost and greater liquidity of US index ETFs, would it make more sense for someone following the Market Compass models just to use the US signals as a timing tool for US ETFs, rather than using the Monthly World model and buying an MSCI World ETF? Given the high correlation between the US stock market and the MSCI World index, and given that fee differences could amount to more than 0.4%, wouldn’t it be simpler to stay closer to home and buy the US ETF?

For an investor looking to follow only one asset class, there are only two components of the Market Compass methodology that matter, namely macroeconomic conditions and trend (the other components that come into play in allocating between asset classes – relative momentum and valuation – don’t matter if there’s only one asset class under consideration).

I compared the effects of applying these two components to index data going back as far as 1970. This test used only US macroeconomic data, as historical data are not as readily available – and because, in practice, Market Compass testing suggests that US macro data provide pretty good guidance as to whether conditions are favourable for investment in other geographies as well (see the note on the Macro Component in the Methodology section for further discussion of this). The rules modeled are fairly simple. Using a broad variety of high frequency, concurrent measures of US economic health (unemployment claims, PMI data, regional fed surveys, national surveys such as the CFNAI), and a simple trend signal (average of 3-, 6- and 12-month price change greater than 0), if any 3 macroeconomic measures plus the trend signal are negative, then the model exits. Otherwise, it remains invested. These are essentially the rules used in the Monthly World Models tracked on the blog.

Historically, a buy-and-hold investor in the US stock market would have outperformed an investor in the World index, although there were periods where the opposite would have been true. The chart below shows the differences in buy-and-hold returns:

Figure 1. Buy-and-Hold Returns for S&P 500 and MSCI World (USD, Gross), 1970-2012

If we apply the Market Compass macro and trend components to the indices, we see improvement in both:

Figure 2. S&P 500 and MSCI World, Model Results vs. Buy-and-Hold

The application of the model delivers a substantial improvement to both, but even more so to the World index than to the S&P 500. This is perhaps consistent with the view that the S&P 500 tends to be more efficiently priced than markets elsewhere in the world. This outperformance does not hold all the time – the S&P 500 caught up with the World during the bull market of the 1990s, for instance – but it seems to kick in at inflection points where the models get in or out of the market. We can see that the improvement extends to risk measures as well as to returns:

Figure 3. Annualized Return and Volatility, S&P 500 and MSCI World

The Market Compass models deliver improvement to both return and risk for both indices, but the improvements to both sets of measures are greater for the MSCI World. We can see this clearly if we combine return and risk to calculate Sharpe ratios:

Figure 4. Sharpe Ratios

Applying the model to the MSCI World delivers a 0.17 point improvement to the Sharpe ratio, vs. a 0.09 gain in the S&P 500.

To return to the original question, this analysis suggests that the MSCI World index may offer  some additional benefit. The annualized outperformance of roughly 0.3% is not enormous, but it is enough that a low-cost global ETF like Vanguard’s Total World Stock ETF (with an MER of 0.19%) would still offer some theoretical return advantage over a US ETF at 0.05-0.06%, based on historical returns. Transaction costs and withholding taxes could eat into the performance advantage of the World approach, but ultimately the difference in returns remains pretty minimal either way.

However, perhaps more important than historical returns is the historical risk picture. A world index is, by nature, more diversified than a US-only ETF, which offers some risk mitigation in and of itself, and the analysis above suggests that the application of Market Compass-type models could improve this risk-reduction benefit even more. We have already seen in a post some time ago that there is diversification benefit in using a 3- or 4-region approach rather than an MSCI World only approach, but for an investor who chooses to put all of their eggs in one ETF basket, it makes sense to the author of Market Compass to make that basket broader rather than narrower.