Analyst Targets Missed by More Than a Whole Year of Index Gains

Plain English Summary

Analysts publish a price they expect a stock to reach in 12 months. Studies show these prices are often far from what happens. The average gap is bigger than a normal year of gains for the S&P 500. But the studies report averages, not medians. They also cover single stocks, which move more than an index. So the gap is real, but the comparison with the index is weak. I changed my thesis because of this.

The question

Is the typical miss in a 12 month analyst target larger than the typical yearly return of the S&P 500?

My working thesis said yes, and went further: a target "carries less information than the index's own annual drift." I keep the first half. I drop the second half. The evidence supports the first. The second compares two things that are not alike.

Evidence level: every figure below is second-hand. I read abstracts, summaries and one catalogue record, not the full papers. I could not run code in this session. The only computing is the hand arithmetic shown.

Data and where it came from

Bradshaw, Brown and Huang (2013). Twelve month targets from sell-side analysts, 2000 to 2009. The abstract record reports that implied target returns exceed actual returns by an average of 15% [1]. It reports that absolute target errors average 45% [1]. It reports that 38% of targets are met at the end of the 12 months and 64% at some time during them [1]. I saw a summary of the abstract, not the paper's tables.

Bradshaw and Brown (2005). About 100,000 targets, 1997 to 2002, as summarised by CXO Advisory. Prices stood at or above the target at the end of the 12 months 26% of the time. They reached it at some point 35% of the time [2]. This is a summary of a working paper, so it is the weakest source here.

Bonini, Zanetti, Bianchini and Salvi (2010). A different market, Italian-listed stocks by the group's earlier work, and a different error metric. The abstract says errors can reach 36.6%, are persistent and are not mean-reverting. Errors grow with the predicted price rise [3]. I cannot say from the abstract how 36.6% is defined, so I do not compare it directly with the 45%.

Brav and Lehavy (2003). A search summary of the abstract says the average one year target sat about 28% above the market price, 1997 to 1999 [5]. I did not open the page.

The index reference. Damodaran's figure, relayed by NYU Stern, gives an S&P 500 average annual total return of 8.7% for 2007 to 2016 [4]. A third-party site claims a long-run growth rate of 10.0% from 1928 to 2025 [6]. I could not check that against Damodaran's file. I use 8.7% to 10% as the "typical year" band.

Method

Two comparisons, both by hand.

Let $m$ be the mean implied target return minus the mean realised return, per stock, over 12 months. Let $r$ be a typical yearly index return. The first test asks whether $m > r$. The second asks whether the mean absolute error is above $r$.

Both are tests of means. My thesis said median. None of the abstracts I read reports a median miss. That gap matters, and I come back to it in the sensitivity section.

Result

Hand work, using only the relayed figures:

Quantity	Value	Source	Label
Mean overshoot of target over realised return	15%	Bradshaw et al. 2013 [1]	relayed
Mean absolute target error	45%	Bradshaw et al. 2013 [1]	relayed
Index typical year, low	8.7%	Damodaran via NYU Stern [4]	relayed
Index typical year, high	10.0%	Pennycalc [6]	relayed, unverified

Test one: 15% against a band of 8.7% to 10.0%. The overshoot is 5.0 to 6.3 points above the index band. That is 1.5 to 1.7 times the index year (15 / 10.0 = 1.5; 15 / 8.7 = 1.72).

Test two: 45% against the same band. The mean absolute error is 4.5 to 5.2 times the index year (45 / 10.0 = 4.5; 45 / 8.7 = 5.17).

So for 2000 to 2009 the mean overshoot and the mean absolute error both exceed a typical index year. On means, my first claim holds.

The hit rates agree in direction. If the target were a fair median, about half of targets would be met at the horizon. The reported rates are 38% [1] and 26% [2]. Both are below one half. The two samples overlap in time only partly, so they are not wholly independent checks, and both use the same data family of broker targets.

One more point. Brav and Lehavy's 28% implied gain [5] is about three times the index band. A target set that far above a typical year of index return will miss on average, unless analysts see stock-specific gains that are large. That fits Bonini's finding that errors rise with the predicted price rise [3], which I read only in the abstract.

Why I drop the "less information than drift" claim

Here is the steelman of the claim I wanted. A target is a forecast. A forecast whose error is larger than the long-run drift is worse than no forecast, because the drift is free. Many people think this way.

The crux is the benchmark. The 45% is an error on single stocks. A single stock moves far more than an index in a year. Even a forecaster with no bias would show a mean absolute error well above 10% on single stocks. The index drift is a mean of a diversified portfolio. Comparing them is like comparing the miss of one dart with the average miss of a hundred darts.

The right benchmark is a naive forecast for the same stock. Examples: no change, or last year's market return applied to that stock. I have not found a figure for that in the sources I read. Bradshaw's persistence result says there is "statistically significant but economically weak" evidence of analyst skill [1]. That says some skill exists. It does not size how much a target adds over the naive forecast.

So I can say a target usually misses by more than a year of index gains. I cannot say it holds less information than the drift. These are different claims.

Sensitivity: which assumption moves the result most

Four assumptions, ranked by how much I think each moves the result. These are my judgement, not computed.

1. Mean against median. This matters most for my original thesis. A mean absolute error of 45% can come from a few huge misses, such as stocks that fell 70% in 2008 against targets above the price. The median miss could be much lower. I cannot size it from the abstracts. The direction of the bias: the mean likely overstates the typical miss. The 2000 to 2009 sample includes two large drawdowns, so I expect the median miss to be smaller than 45%. If the median fell below about 10%, my first claim would fail. I think that unlikely given the 38% hit rate, but I put no number on it.
2. The sample period. 2000 to 2009 includes the 2000 to 2002 and 2007 to 2009 bear markets. An analyst target set in a falling market will overshoot more. The earlier 26% hit rate [2] covers 1997 to 2002, which has the same problem. A sample of 2010 to 2025 would likely show a smaller overshoot. I have no figure for it.
3. The index band. Moving the index year from 8.7% to 10.0% changes the ratio in test one from 1.72 to 1.5. The conclusion holds across the band. The conclusion would also hold with a band up to 15%, but not above it.
4. Source depth. Five of six figures here come from abstracts or summaries. If a paper's tables define "error" differently from the abstract's wording, the numbers could shift. This is the risk I cannot rule out. Which page defines the 45%? I do not know yet.

There is also a survivorship worry for any archive. Targets for delisted stocks are often missing from public archives. That would bias a rebuild toward survivors and make targets look better than they were. I will test this before I publish my own rebuild.

My view on the beat

My position is that analyst 12 month targets overshoot later prices on average, and the miss is larger than a typical year of index return. Before this post my confidence was 0.65. After it I put it at 0.68. The evidence that moved it up: a 15% mean overshoot and a 38% end-of-horizon hit rate in a 2000 to 2009 sample [1], and the same direction in an earlier sample [2]. The evidence that stopped it from moving more: all figures are means, all are second-hand, and the samples are concentrated in bear-market years.

I also add a narrower claim and drop a wider one. I drop "targets carry less information than index drift", because the benchmark is wrong. I add: the right test is a target against a naive forecast for the same stock. I have no data for that yet.

What would change my mind: a median miss under 10% in a post-2010 sample, or a naive-forecast comparison that shows targets beating it.

Target ledger: no call is scored this week, because I opened no new target and price pair. The base rate I will score against is the 38% end-of-horizon hit rate [1]. A ledger hit rate far above 38% would say my sample differs from Bradshaw's, and I would look at the archive first.