Siegel vs. Zweig: What are long-run stock returns?

I see that Jeremy Siegel replied to Jason Zweig’s criticism that Siegel’s studies of stock returns contain sample size biases. One of Zweig's criticisms is that for the early periods, e.g. 1815 and 1834, Siegel’s data is composed of only a few stocks. The results biased the returns upward.

Siegel responded that while there are problems with the early data, he pointed to Goetzmann and Ibbotson’s work, A New Historical Database for the NYSE 1815 to 1925: Performance and Predictability, as being free from survivorship bias and supportive of his conclusions about long-term stock returns of about 7% per annum.


But there is survivorship bias in the data!
All this bickering about the data still doesn’t make sense to me. Let’s do a sanity check, as I did in my post What actually happens in the long run:

What if your family had managed to save the equivalent of $100 at the time of Augustus Caesar (give or take 2,000 years ago) and put it into equities or an equivalent investment? At 7% a year, the value of your family’s $100 original investment would now have 60 zeros behind it. Your family could finance TARP and the bailout by the world’s central banks from the chump change derived one day’s interest.

What happened?

What happened was in the intervening 2,000 years, there were many upheavals that destroyed wealth. Empires fell, starting with the Roman Empire, barbarians sacked cities and a lot of people died in very unpleasant ways.

Quantitative analysis without context
Children draw pictures by coloring the numbers. This is an example of quantitative analysis by coloring the numbers. Proper analysis needs proper context. Many of these quants are far smarter than me, but sometimes they need to step back and really, really think about the assumptions behind their models.

Here is an example from David Halberstam’s The Best and the Brightest about Robert McNamara, who I consider to exemplify the greatest quant failure of our time:

[His] mind was mathematical, bringing order and reason out of chaos. Always reason. And reason supported by facts, by statistics — he could prove his rationality with facts, intimidate others. He was marvelous with charts and statistics.

Once, sitting a CINCPAC for 8 hours watching hundreds and hundreds of slides flashed across the screen showing what was in the pipe line to Vietnam and what was already there, he finally said, after 7 hours, “Stop the projector. This slide, number 869, contradicts slide 11.”" Slide 11 was flashed back and he was right, they did contradict each other.

Everyone was impressed, and many a little frightened. No wonder his reputation grew; others were in awe. For it was a mind that could continue to summon its own mathematical kind of sanity into bureaucratic battle, long after the others, the good liberal social scientists who had never gone beyond their original logarithms, had trailed off into the dust.

Though finally, when the mathematical version of sanity did not work out, when it turned out that the computer had not fed back the right answers and had underestimated those funny little far-off men in their raggedy pajamas, he would be stricken with a profound sense of failure, and he would be, at least briefly, a shattered man.