Technical Analysis vs Quantitative Analysis: A Rule Tested Decade by Decade

Michael Whitman · Trading since 2017, full-time since 2020 · YouTube creator, 137K subscribers · Published 2026-09-09 · Updated 2026-10-27 · Reviewed 2026-10-27

Technical analysis reads a chart through patterns, levels and indicators judged by a person; quantitative analysis writes the idea as a rule, runs it on past data and checks it on data it was not built from. On the S&P 500, all 25 moving-average rules tested here had a positive gap in the 1960s, 1970s and 1980s, and none of the 25 did in the 2010s or 2020-2026.

Both approaches work from the same price history. Run on seventy years of the index, a simple chart rule passed the first kind of test convincingly and failed the second, and choosing it again every ten years did not rescue it.

What each one is

Technical analysis reads charts through patterns, levels, indicators and trend, interpreted by a person. Technical analysis covers the tradition, and price action covers its most judgment-heavy end.

Quantitative analysis writes a rule precisely enough to run mechanically, measures what happened, and keeps part of the history back to check the result on. Quantitative analysis covers the method.

One produces readings and the other produces claims. A chart reading describes what someone sees. A quantitative claim is a specification that anyone can run, get the same numbers from, and disagree with.

Where they differ

Whether the claim can be rerun. Both start from the same chart, yet a chart reading depends on which level, which pattern and how strictly each was judged, so two people can see different things in one chart. A written rule gives the same answer every time it is run, which is what makes disagreement useful.

How each one guards against luck. A chart reader’s sense that a pattern is reliable comes from the cases remembered. A quantitative test can compare a result with what shuffled data produces, and with what the rule did on a later period.

What each one adapts to. A person can weigh something that has never happened before, while a rule cannot, and will apply itself to conditions nobody had in mind when it was written.

How each one is usually shown. Chart lessons are mostly marked up after the outcome, whereas a quantitative result is supposed to arrive with its in-sample and out-of-sample numbers side by side.

The test, written as a rule

The chart idea: “the market behaves better above its moving average.” Written as a rule: on each day, note whether the S&P 500 closed above its N-day simple average, then compare the average next-day change after closes above with the average after closes at or below. That gap is the statistic, in percent per day, for 25 lengths: 10, 20 and so on up to 250 days.

The split. Every measured day from 26 Feb 1958 to 1 Oct 2026 has all 25 averages available. The first half, to 17 Jun 1992 (8,632 days), is where a researcher would pick the best length. The second half, 18 Jun 1992 onward (8,632 days), is the check.

The shuffle. To see what the best of 25 rules looks like on data with no order in it, the first half’s daily changes were put in random order 1,000 times, a price path was rebuilt from each, and the best of the 25 gaps was recorded.

A rule that passed one test and failed the other

In the first half, every one of the 25 lengths showed a positive gap. The best was the 10-day average: next-day changes after closes above it averaged 0.1221% a day more than after closes below.

The shuffles never matched it. Across 1,000 reordered histories, the median best-of-25 gap was 0.0189% and the 90th percentile 0.0385%; none reached 0.1221%. By that test the first-half result was not luck. It also shows how easily noise flatters a search: in 87.6% of the shuffles, the best of 25 lengths still came out above zero.

In the second half the same 10-day rule ranked 25th of 25, with a gap of -0.0593% a day. Only three lengths (220, 230 and 240 days) stayed above zero, the best at 0.0049%. The rank correlation between the two halves’ 25 results was -0.735: the lengths that looked best early tended to look worst later.

Re-choosing the rule every decade

A fair objection to one split is that nobody would keep a 1992 choice for 34 years. So the same 25 gaps were measured decade by decade, from 1960 to 23 Oct 2026, and the best length of each decade was checked in the decade after it, the way a researcher who re-tests every ten years would use it.

The pick usually ranked well, and still often lost. Of six picks, five ranked in the top half of the next decade (1st, 10th, 4th, 3rd, 13th and 2nd of 25), but three of the six had a gap at or below zero there. Picking well among 25 rules is a different thing from picking one whose result is above zero.

What changed was the whole set, not the choice. All 25 lengths were positive in each of the 1960s, 1970s and 1980s; none was in the 1990s, 14 were in the 2000s, and none was in the 2010s or in 2020-2026.

Which one to use: Technical Analysis or Quantitative Analysis?

Use a quantitative test when the question is whether something held up, and insist on the part of the test run on data the rule was not chosen from. The first half’s result above was real for its period and still did not carry forward.

Use judgment rather than a backtest when the situation is new, such as an event no rule was specified for. A person who sees that conditions have changed is doing work no backtest does.

Use both by writing your chart reading down as a rule. The exact level, the exact condition and what would disprove it turn a reading into something that can be run, which anyone willing to be that precise can do.

And when a pattern looks obvious, test it on a later period before trusting it. The obviousness comes from the data you already looked at.

Why the rule turned around

Next-day persistence on the S&P 500 changed sign over the decades. The correlation between one day’s change and the next was 0.1532 in the 1960s and 0.2487 in the 1970s, so up days tended to follow up days, which is what a short moving-average rule picks up. It fell to 0.0531 in the 1980s and 0.0171 in the 1990s, and turned negative: -0.0878 in the 2000s, -0.0471 in the 2010s and -0.1661 from 2020 to 2026.

So the market the rule was found in was not the market it was applied to. A shuffle test asks whether a result could come from the same numbers in random order. It cannot ask whether the process producing the numbers will stay the same, which is the question only a later period answers.

The original data

Sample: S&P 500 daily closes from 4 Mar 1957, the day the 500-stock index began, to 2 Oct 2026 (Yahoo Finance, ^GSPC). Measured days need 250 closes before them for the longest average, so the test starts on 26 Feb 1958; the last measured day is 1 Oct 2026, whose next-day change runs to 2 Oct.

Shuffle test: 1,000 random reorderings of the 1958-1992 daily changes (fixed seed, so it reruns exactly); best-of-25 gap median 0.0189%, 90th percentile 0.0385%, highest 0.0605%, so none of the 1,000 reached the real 0.1221%.

Next-day correlation by decade: 1957-59 0.0967 (714 pairs), 1960s 0.1532, 1970s 0.2487, 1980s 0.0531, 1990s 0.0171, 2000s -0.0878, 2010s -0.0471, 2020-2026 -0.1661; over the whole span -0.0079. The pair counts are in the S&P 500 next-day correlation file.

The decade test (new download, 27 Oct 2026). The same rule on ^GSPC closes to 23 Oct 2026, grouped by the calendar decade of the measured day; 2020-2026 runs to 22 Oct 2026 and has 1,711 measured days, the full decades 2,489 to 2,528. A pick’s rank counts the other 24 lengths with a larger gap, so 1st is the best of 25. The best length of 2020-2026 so far is the 100-day, at -0.0020% a day. Every break of more than four calendar days in the file is a known market closure, such as September 2001 and October 2012, not a missing row.

Second source. FRED’s S&P 500 series (27 Oct 2016 to 23 Oct 2026) gives the same 1,711 measured days for 2020-2026, the same best length (100 days), the same 2nd place for the 2010s pick and a rank correlation of 0.998 with the Yahoo gaps, which it never misses by more than 0.002 points. Its 96 blank dates are market holidays, on which Yahoo has no row either; 2,503 of the 2,511 shared closes agree within $0.01, the largest gap $5.29 on 11 Aug 2021. Every decade and length is in the decade-by-length file, and the check in the FRED check file.

What this does not show. The gaps are averages of next-day changes before any cost; nothing here is a trading result, and none of the 25 rules is offered as one. All 25 lengths, both halves, are in the S&P 500 moving-average split file.

When it fails

The characteristic technical failure is confusing familiarity with evidence. Someone who has looked at charts for years has seen a setup resolve well many times and remembers those cases better than the ones where nothing happened. The confidence is genuine and the sample was chosen by memory. Counting is the remedy, and the count has to include the cases that were forgotten.

The characteristic quantitative failure is stopping at the in-sample result. The 10-day rule above had the best first-half number and a result no shuffle matched, and it was the worst of the 25 afterward.

A third is testing many variations and reporting the winner. The best of 25 beat zero in 87.6% of shuffled histories.

A fourth is re-testing often and mistaking a good rank for a good result. The decade picks ranked well and were still at or below zero half the time.

And a fifth is treating either as a complete method, since neither sets a position size or a loss limit by itself.

Questions people ask

What is the difference between technical and quantitative analysis?

Technical analysis is a person reading a chart; quantitative analysis is a written rule run on data and checked on a later period. The same moving-average idea appears in both, and only the written version produced the numbers on this page.

Is quantitative analysis better than technical analysis?

It is better at one job: saying whether an idea held up. On the S&P 500 it showed a rule that looked strong from 1958 to 1992 ranked last of 25 afterward, which a chart reading alone could not have shown.

Can technical analysis be backtested?

Only the part that can be written as a rule. Here “closed above its N-day average” became 25 exact rules, measured over 1,711 to 2,528 days a decade.

Why do backtested rules stop working?

Because the market behind the numbers changes. Next-day correlation on the index went from 0.2487 in the 1970s to -0.1661 in 2020-2026, and the count of positive lengths fell from 25 of 25 to none.

Related

Technical analysis covers the interpretive tradition on the left of this comparison. Quantitative analysis covers writing and testing a rule, including the split used here. And price action covers the most judgment-heavy way of reading the same charts.

This page is educational, not financial advice. Test every idea on your own charts before risking money.