Old signal vs new: the calls compared, and what we're testing next

On October 13 the standard Buy, Hold and Sell changed: from half the plays and half analysts' upside to the plays, an exit rule (Off the high) and the trend (what changed and why). The case for it was a test of how often each signal's Buys beat the market. This piece tests the change harder: back to 2006, on how badly things went as well as how often, and against the old signal itself, rebuilt from analysts' price targets for the stretch where we have them. Not all of it flatters the new signal.

Tonight's calls, old and new

On the October 13 close, the two signals agree on 127 of the 375 names both can score (34%). The old signal called 46 stocks a Buy that the new one calls a Sell, and 4 a Sell that it now calls a Buy.

Rows are the old standard (plays 50, analysts' upside 50), columns the new one, on tonight's numbers. The names that moved most are listed with the change.

Since 2006: how often, and how badly

At every month-end since 2006, each covered stock was scored exactly as the signal would have scored it then, from prices and the plays' calls up to that close. Each Buy and Sell was then checked against the stock's market (the S&P 500, or the TSX for Canadian names) over the next year, and against how far the stock fell in that year. "Plays only" is the half of the old signal that can be tested this far back. "Fell 30%+" means the stock was 30% or more below the day of the call at some point in the following year; "flips a year" is how often a stock's call changed, per stock per year.

• The exit rule does its main job. A stock the new signal called Sell was twice as likely as a Buy to fall 30% or more at some point in the following year (33% against 16%). The worst tenth of its Sells trailed the market by 41% or more over the year; the worst tenth of its Buys by 28% or more.
• How often it's right barely moved. The edge, how much more often Buys beat the market than Sells, is +3.9 pts against +2.9 pts for the plays alone. Many Sells still beat the market: beaten-down stocks often bounce, and the ones that went to zero aren't in today's list to count against them.
• It changes its mind less. 4.3 changes per stock per year, against 6.5 for the plays alone.

It depends on the market

The numbers, year by year

Calls made in	Edge over the next year
2006	−6.9 pts
2007	+3.5 pts
2008	−7.5 pts
2009	−2.5 pts
2010	+8.5 pts
2011	+19.8 pts
2012	+19.4 pts
2013	+17.4 pts
2014	+22.8 pts
2015	+7.8 pts
2016	−4.1 pts
2017	+13.4 pts
2018	+6.5 pts
2019	+18.4 pts
2020	−20.5 pts
2021	+15.9 pts
2022	−9.9 pts
2023	+3.0 pts
2024	+2.6 pts
2025	−5.5 pts

Calls made in 2008 (−7.5 pts), 2020 (−20.5 pts), 2022 (−9.9 pts) and 2025 (−5.5 pts) did badly: in each of those years a sharp, broad fall was followed by a fast rebound. An exit rule turns stocks to Sell after they've fallen, so when the whole market falls and snaps back, it sells close to the bottom and misses the bounce. In years with a steadier trend it did well. Since 2019 its edge on how often it's right has been about zero, though its Sells were still the ones more likely to fall hard (31% against 19% for Buys).

The old signal, rebuilt

There's no stored history of analysts' consensus targets, which is why the old signal couldn't be tested. But rating changes with price targets are on file from late 2023, so from July 2024 we rebuilt it: each firm's latest target in the past year, averaged over three or more firms, for 188 stocks. It's an approximation: on tonight's numbers the rebuilt upside and the published consensus agree closely (correlation 0.68, a median gap of 7%), but not exactly. Year-ahead results run to calls made in September 2025; three-month results to September 2026.

In this stretch the old signal did better. Its Buys beat its Sells over the following year; the new signal's didn't. Over three months the two were close (edges of +2.7 pts and +2.1 pts).

The clearest test is where they disagreed. 318 times a stock was a Buy under the old signal and a Sell under the new one, typically 39% below its 1-year high. Those stocks beat their market over the next year 59% of the time, by a median +12%. The reverse case (679 times, old Sell and new Buy) beat it 52% of the time, by a median +3%.

That stretch had one sharp V: the April 2025 tariff sell-off and its rebound, the kind of market where an exit rule costs the most and where analysts' targets, which barely moved, pointed at the recovery. It's also only about fifteen months of year-ahead results, with no long bear market in it. It's a real result, not a verdict.

Stocks already 20% off their high

This is where the two disagree most. Since 2006, the new signal called 90% of these Sell and almost none Buy; the plays alone called 75% Sell. Over the next year 38% of those Sells fell another 30% or more at some point, and 55% still beat their market.

From July 2024, the old signal called 13% of stocks this far down a Buy. Those Buys beat their market 60% of the time, a median +14%, but 37% fell another 30% or more first. Buying the fall paid on the typical stock and hurt badly on a third of them; the exit rule gives up some of the first to avoid the second.

What we're testing next

Each idea below was scored the same way, month by month since 2006. "Crash years" is the average edge for calls made in 2008, 2020 and 2022.

• A market guard looks best. When the stock's whole market is 15% or more below its own high, the exit rule stands down and the plays decide. It lifts the edge to +4.7 pts by doing less damage in 2008, 2020 and 2022, and changes nothing in calmer years.
• Sticky calls cut the noise. If a Buy stays a Buy until the score falls below 50 (and a Sell until it rises above 50), signals change 2.8 times a year instead of 4.5 with the guard on, for −0.5 pts of edge. Fewer flips means fewer alerts that reverse a month later.
• Analysts' upside, capped, at a small weight was the best of the additions in the rebuilt stretch over three months (+3.2 pts, against +2.1 pts for the new signal), but two years is too short to put it back. Analysts' target revisions (raised or cut in the last 90 days) added nothing (+2.2 pts). We keep collecting targets so both can be graded properly.
• Small or mixed: moving the exit rule's floor to 20% or 40%, scaling it to each stock's volatility, using 12-month momentum (skipping the latest month) for the trend, adding return on equity and debt, or asking for a 70 to Buy while the S&P 500 is below its 200-day average each land within half a point of today's standard over the whole period; some do better in one stretch and worse in another. Measuring the fall against the market's own fall raised the overall edge but made the crash years worse.

Two things would make every one of these tests more honest, and both are on the list: stocks that no longer trade (the test uses today's list, which flatters buying dips and hides the cases an exit rule exists for), and a longer record of analysts' targets and of companies' guidance, which the site has started keeping. Any change to the standard will be published here with its record first.

The fine print

Study run 2026-10-13 on 349 covered stocks, month-ends 2006 to September 2026, with each stock scored from data public at the time. "Beat the market" compares the stock's return over the next year with its market's. Edges of a few points are normal and noisy; a year can swing by twenty. The rebuilt old signal uses targets that aren't split-adjusted, so any target more than double or less than half the price on its day is dropped. None of this knows anything about an individual account, and a better signal doesn't make one stock a safe place for most of one (one stock too many).