Tactical country allocation

How this differs from the Forecaster

The two tools answer different questions and will regularly point in opposite directions. That is by design and not a fault in either.

A market can sit at the top of the Forecaster and the bottom of this page at the same time. Indonesia is the clearest current example: cheap enough on CAPE to carry one of the highest long-run forecasts, and simultaneously among the worst performers of the past year and below its own trend. Both readings are correct. Cheap markets are usually cheap because something is going wrong, and the thing going wrong often continues for a while.

The three signals

Momentum, 70 percent of the score

Total return over the 12 months ending one month ago. The most recent month is deliberately skipped: short-horizon returns tend to reverse, so including it adds noise and, in individual securities, actively works against you. Markets are then ranked against each other rather than scored in absolute terms, because what matters for choosing between countries is the ordering.

Trend, as a gate rather than a score

A market is eligible only if its latest monthly close sits above its own 10-month moving average. This is applied as a filter rather than blended into the score, and the distinction matters: if it were part of the score, a market with spectacular momentum could outvote its own broken trend, which is exactly the situation the filter exists to prevent.

When most markets fall below their averages at once, the rule holds very little or nothing. An empty list is a real answer, not a failure to produce one.

Value, the remaining 30 percent

Where each market's CAPE sits in its own history, the same measure the Forecaster uses, inverted so that cheap ranks high. Momentum gets the larger weight because the country-level evidence for it is stronger, but the two are worth combining: Zaremba found value and momentum reinforce each other in double-sorted portfolios rather than measuring the same thing twice. The weighting is adjustable on the page itself.

Earnings revisions, accruing but not yet scored

A fourth signal is being collected and does not yet feed the ranking. Changes in analysts' forward earnings estimates are one of the better-documented country signals, and the usual way to get them is to buy an I/B/E/S feed. They can instead be derived from the factsheets already downloaded each month, because MSCI define forward P/E as price divided by the rolling 12-month consensus estimate:

Both terms are printed on the factsheet, so storing each month's snapshot produces the signal for nothing. The catch is that it cannot be backfilled. MSCI serve only the current month at their public URLs, and the Internet Archive holds seven scattered snapshots of the World factsheet and fewer of the individual countries. The record therefore begins the first time it was stored and thickens by a month at a time. Until two months exist the column shows a dash rather than a zero, because no reading is different from no change.

What the evidence actually says

Country momentum is among the better-documented effects in asset pricing, and the headline figures are considerably better than what an investor should expect to capture.

The gap between ten points and five is the important part of that table. Index studies assume you can hold an index; the ETF study measures what was left after the spread, the tracking error and the fees. Five points a year of excess return would still be an excellent outcome, and it is the number to anchor on.

What our own rule actually did

Citing other people's results is not the same as checking your own, so this rule was run on the same ETF prices the page uses, walking forward month by month with every signal computed only from data available at the time. The result is far more modest than the published figures, and worse than a plain global index over the last decade.

The last ten years

Over twenty-one years

Extending back to 2005 changes the picture, because the period then contains a genuine crisis. Against an equal-weighted basket of the same 38 markets the rule beat it by 2.1 points before costs and 1.2 after, and the trend filter earned its keep:

The filter roughly halved the worst loss, from about 58 percent to about 35. That is the whole of its contribution and it arrives in one concentrated dose during a crash. Over the last ten years, which contained no comparable event, momentum without the filter had a marginally better Sharpe ratio and less turnover. The filter is insurance: it costs something every month and pays once a decade.

Against a global tracker, over matched windows

Global index funds cannot be compared over the full period because they did not exist for all of it. The ACWI fund launched in 2008 and the World fund in 2012, so each is set against the rule over exactly the months it covers.

Most of the gap against a tracker is not a failure of the signal. A cap-weighted global fund has been roughly two thirds American through the best stretch American equities have ever had, while this rule treats the United States as one of 38 candidates. Judged against its own universe, which is the fair test of whether the ranking works, the signal adds value consistently. Judged as an alternative to owning the world, it has not been one.

What the test cannot show

The value tilt is untested. Research Affiliates publish where CAPE sits today, not a history of where it sat, so the backtest covers momentum and trend only. That is 70 percent of the live rule's weight, and whether the remaining 30 percent helps is unknown.

The universe was chosen with hindsight. The ETF list is today's list, so funds that launched and closed are missing and every market enters from its own fund's inception. The test knows which markets stayed investable, which flatters a concentrated strategy more than it flatters the benchmark.

Costs are modelled rather than measured. Turnover was computed exactly and averaged 38 percent a month, charged at 20 basis points round trip. Real spreads on the thinner funds are wider than that, and no tax is included at all.

Concentration is noise. Holding the top four rather than the top eight was worse over ten years and better over twenty-one, which is a sign that the difference between them is not information.

Is there a rule that beats owning the world?

The country momentum research benchmarks against equal-weighted country indices, so this rule beating equal weight is the published finding reproduced. Nobody claimed it would beat a cap-weighted global fund. That is a different question and worth asking separately, because owning a global tracker is what most people should compare against.

The most direct candidate is not country rotation at all. It is to hold the tracker itself and step aside when it falls below its own trend. Running that on the same prices gives a result that depends almost entirely on whether the window contains a bear market.

The global funds have only existed since 2008 and 2012, so their whole record is post-crisis. Over that stretch an overlay sits out about a fifth of the months, misses upside and loses on both return and Sharpe ratio. The US series reaches back to 1997 and contains the dot-com collapse and the financial crisis, and there the picture reverses: absolute momentum beat buy and hold on return, on Sharpe ratio and on drawdown, cutting the worst loss from 51 percent to 19.

What actually happened to people who bought this

The most useful evidence is not a backtest at all. Tactical allocation funds have existed for decades and their record is measurable. Over January 1994 to October 2016 they underperformed every benchmark index tested, on both absolute and risk-adjusted measures. Equally weighted portfolios of them trailed matching portfolios of index ETFs by between 1.77 and 5.15 percentage points a year, a gap well beyond their fees. Seventy percent did worse than a passive balanced fund since their own inception, by an average of 2.6 points a year, and they showed significantly negative alpha throughout, including through 2008 when the whole proposition should have paid off.

That is professionals with research budgets running tactical strategies in real money, and it is the number to weigh against any backtest on this page, including ours.

Where the numbers come from

Prices are in US dollars, so momentum includes the currency move. That is the right choice here rather than an oversight: rotating between country ETFs, the currency is part of your return whether you wanted the exposure or not.

Only completed months are used. A signal computed part-way through a month would change depending on which day you looked at it, which is not a useful property for a rule that rebalances monthly.

What it costs, honestly

Every figure quoted on this page is gross. A monthly rebalance across country ETFs means twelve sets of dealing costs and spreads a year, and outside a tax shelter it means realising gains on a schedule decided by a formula rather than by you. For a UK investor in an ISA or SIPP that is manageable; in a taxable account it can consume the entire edge.

There is also the behavioural cost, which is the one that actually stops people. Trend following sells after markets have fallen and buys after they have risen. It will have you selling a market the month after a crash and buying back higher. Anyone who cannot follow that without second-guessing it will get the worst of both approaches.

What this cannot do

It cannot tell you whether to use a tactical overlay at all. That is a question about your temperament, your tax position and how much tracking error against a simple global fund you can tolerate, none of which appear in the data.

It cannot forecast returns. The Forecaster attempts that and states its record. This page makes no return forecast whatsoever; it reports what prices have done and applies a rule to them.

It does not account for your existing holdings, your currency, your costs or your tax. And the historical results behind it come from backtests with the benefit of hindsight about which markets survived and remained investable, which flatters every strategy of this kind.