Why Market Timing Is Unreliable Even When Some Signals Work

We took the most popular timing rule in circulation, wrote it down before testing it, and ran it against staying invested across every fifteen-year stretch of the Indian market — with tax, with costs, and against a control designed to catch us fooling ourselves. The signal turned out to be real. It still lost every single time.

Updated 3 September 2026

SIP vs Lumpsum CalculatorOpen

The rule we tested, written down first

Timing usually gets argued in the abstract, which is why the argument never ends. So we tested something specific:

Sell when the index closes below its 200-day average. Buy back when it closes above.

We chose it because it is the one actually promoted, not because it flatters our conclusion.

Before running anything we wrote down the rule, what it would be compared against, and what would count as it working — including that we would change this article's title if it did. That document went into our repository before a line of the test existed. The rest of this article asks you to be suspicious of timing evidence, and that request is worth nothing if we did not hold ourselves to it.

How it was tested

Every fifteen-year period in our data, starting one month apart — 136 periods from 2 May 2000 to 3 August 2026. Every approach starts on the same day with the same money.

  • Stayed invested. Bought once, held throughout.
  • The 200-day rule. In and out as the rule dictates, decided on the previous day's close.
  • Traded just as often, on random dates. The control, explained below.
  • Followed the rule backwards. Out when the rule says in.
  • Held a fixed part in cash and never traded. Added after we saw the results, and flagged wherever it appears.

Every sale pays capital gains tax in the financial year it happens, at Indian rates for listed equity, with the annual exemption applied once a year rather than once a sale. Cash earns interest, taxed too, and every trade is charged a fee. That matters more than it sounds: someone who stays invested defers tax until they sell, so the government's share keeps compounding for them meanwhile.

What happened

Stayed invested throughoutHeld cash, never tradedFollowed the 200-day ruleTraded as often, on random dates (the c…Followed the rule backwards+2.35%+10.33%+18.30%Return a year after tax and costs, across every fifteen-year period
Each bar spans the middle eight out of ten periods, with the mark on the typical one. The rule and its control sit on top of each other: trading that often produced this result whether or not the dates were chosen by the signal.
Show these numbers as a table
StrategyPoor outcomeTypicalGood outcome
Stayed invested throughout+10.24%+12.38%+16.77%
Held cash, never traded+9.06%+10.90%+14.89%
Followed the 200-day rule+6.58%+8.59%+11.61%
Traded as often, on random dates (the control)+6.84%+8.46%+11.05%
Followed the rule backwards+4.04%+5.15%+6.11%

The rule beat staying invested in 0 of 136 periods. Not a minority — none of them. The typical period returned +8.59% a year against +12.38%, a gap of 4.31% every year for fifteen years. Almost any strategy wins sometimes. This one did not.

"But your trading costs are made up"

They are, and we will not pretend otherwise. We could not find sourced Indian execution costs for an index fund, and inventing a plausible one is exactly what this article warns you about. So instead of choosing a number we charged four.

If a round trip cost youThe rule returnedStaying invested returnedPeriods the rule won
Nothing at all+9.60%+12.40%0 of 136
10 basis points+9.17%+12.40%0 of 136
25 basis points+8.59%+12.38%0 of 136
50 basis points+7.67%+12.37%0 of 136
We could not find sourced Indian trading costs for an index fund, so rather than invent a figure the rule is charged four different ones and you can see the whole range. The top row is the important one: even with trading completely free, the rule loses in every period tested.

Read the top row. With trading completely free — no brokerage, no spread, nothing — the rule still lost by 3.36% a year, and still won in 0 of 136 periods. The conclusion does not depend on the number we could not source, which is the only reason we are comfortable stating it.

The control, and why it is the important part

Here is where most timing articles stop, and where they go wrong. "The rule lost" invites an obvious reply: of course it did, it was a bad rule. Answering that means knowing whether the signal contributed anything at all.

Our first instinct was to run the rule backwards. That is a trap. The index sits above its 200-day average most of the time, so a backwards rule is out of the market most of the time — and it duly did badly, returning +5.15% a year. But it did badly because it was not invested, not because the signal was worthless. It cannot tell those apart, which is a control's only job. Had we stopped there we would have concluded the signal works, and published it.

So we built something better. In each period we recorded exactly what the rule did — how often it switched, how many days it held equity, how long each spell lasted — then generated 1,000 schedules doing all the same things on dates picked at random. Same trading, same time in the market, minus the signal.

1713940no better than chance16%46%76%Where the rule finished among random schedules that traded identically
In each period the rule is ranked against 1000 schedules that switch the same number of times, spend the same days invested, and differ only in choosing their dates at random. The dashed line is the halfway mark. A signal carrying real information about future returns would push these bars to the right.
Show these numbers as a table
Where the rule finished among random schedules that traded identicallyNumber of periodsShare
16% to 19%53.7%
19% to 22%85.9%
22% to 25%96.6%
25% to 29%96.6%
29% to 32%21.5%
35% to 38%21.5%
38% to 41%10.7%
41% to 44%64.4%
44% to 47%42.9%
47% to 50%53.7%
50% to 54%1712.5%
54% to 57%128.8%
57% to 60%64.4%
60% to 63%1712.5%
63% to 66%1611.8%
66% to 69%118.1%
69% to 72%42.9%
72% to 76%21.5%

The rule finished ahead of 54.2% of those random schedules. The middle of the crowd. It stood clear of them — beating 95 in 100 — in 0 of 136 periods.

Which gives the three numbers that explain this article:

  • Staying invested: +12.38% a year
  • Trading that often on random dates: +8.46% a year
  • Trading that often using the signal: +8.59% a year

The trading costs you three and a half points a year. The signal earns back a fraction of one. Nearly everything the rule did to your money, it would have done by throwing darts at a calendar.

Except the signal is not worthless

This is the part we did not expect. The rule's appeal was never really higher returns — it was avoiding the sickening drops, so we measured those too.

Stayed invested throughoutTraded as often, on random dates (the c…Held cash, never tradedFollowed the 200-day rule22.0%40.7%59.5%Worst fall from a high point, across every fifteen-year period
The same periods, measured by the worst drop rather than the return. Here the rule separates from its control rather than sitting on top of it — the one place in this article where the signal demonstrably did something.
Show these numbers as a table
StrategyMildest periodTypical worst fallDeepest period
Stayed invested throughout38.3%59.5%59.5%
Traded as often, on random dates (the control)38.3%53.8%56.3%
Held cash, never traded32.4%50.5%55.4%
Followed the 200-day rule22.1%32.4%34.0%

Staying invested meant living through a worst fall of 59.5%. The rule's was 32.4%: typically 25.5 percentage points shallower, in 136 of 136 periods.

And unlike the returns, this is not explained by the trading. Random schedules that traded identically had a typical worst fall of 53.8%. Against them the rule finished ahead of 95.4% — not the middle — and beat the typical random schedule in 136 of 136 periods.

So the honest finding is not "the signal is noise". It genuinely knows something: it knows when a sustained fall is underway, and it gets you out. It simply cannot turn that into money, because leaving a fall also means missing the recovery, and paying tax and costs both ways.

A signal can be real and still not be worth following. That is the whole article, and it is why the honest version of this argument is harder than "timing doesn't work".

Could you get the same shelter more cheaply?

An obvious question once you see that result, and one we had not planned for. We added this test after seeing the others, which makes it weaker evidence than everything above — a test invented after looking at results deserves more suspicion, including when we are the ones running it.

The rule held equity about 73.8% of the time. So what if you put 73.8% in the index on day one, kept the rest in the bank, and never traded?

That returned +10.90% a year with a worst fall of 50.5%. Against the rule it returned more in 136 of 136 periods — 2.74% a year better — but fell further: the rule had the shallower drop in 136 of 136.

So no. The rule's protection is real and you cannot get it by owning less. Its price is 2.74% a year against the lazy version and 4.31% against staying invested. Whether that is worth paying is a question about you rather than about the data — but you should know you are paying it.

How to judge a timing method someone shows you

Most published timing results fail at least two of these:

  1. Was the rule written down before the results were seen? One adjusted afterwards is not a rule.
  2. Was only information available at the time used? Economic figures get revised and index membership changes; testing with knowledge nobody had that day hands the rule the answers.
  3. Are the false alarms included? Every occasion it said "sell" and was wrong belongs in the result.
  4. Are costs and tax charged? This is a method whose whole approach is to trade.
  5. Is there a control? "We compared it to doing nothing" is not one. Ask what the rule produced with the signal removed and everything else kept.
  6. Is there a stated point at which it would be abandoned? A method with no conditions for being wrong cannot be tested.

Point five is the one nearly everybody skips, ourselves included on the first attempt at this.

The old argument, and what it is good for

Timing is often dismissed with a table like this one.

Return a year₹100 grew to
Stayed invested throughout+13.21%₹2,911
Sat out the 5 best days+11.25%₹1,813
Sat out the 10 best days+9.94%₹1,313
Sat out the 20 best days+7.66%₹743
Sat out the 30 best days+5.65%₹445
Sat out the 50 best days+2.33%₹187
The whole record, recalculated with the single best days removed. This is not a prediction that anyone would miss exactly those days. It shows how much of the total return arrives in a handful of sessions, and therefore how expensive it is to be out of the market at the wrong moment.

Staying invested returned +13.21% a year; missing merely the best 20 days out of 6,759 drops that to +7.66%.

It is striking and routinely oversold. It does not show that people who try to time the market miss those days — nobody's rule targets them. It shows how concentrated returns are, which is a fact about their shape rather than evidence about timing. We have kept it in its proper place.

What this test cannot tell you

One rule, one market, one stretch of history. A market that drifted sideways for a decade could give a different answer.

One version of the rule. Ours switches whenever the price crosses the average, about 7.5 times a year; practitioners often add a buffer to trade less. That would narrow the gap against staying invested, but not rescue the signal — our control traded exactly as often and still matched it.

The trading cost is an assumption, shown as a range rather than picked, and the finding survives at zero.

Some tax detail is left out — surcharge and cess, the section 87A rebate, carrying losses forward. Where we approximated, we did so against the rule rather than in its favour.

It assumes you follow the rule through every stretch where it loses to doing nothing.

What to take away

The question was never whether a signal can work. This one does, in the narrow sense that it sees falls coming better than chance.

The question is whether you can specify it in advance, follow it for a decade including the years it loses, absorb its false alarms and pay its costs and taxes — and still finish ahead of having done nothing. Across 136 fifteen-year periods, the answer was no, every time.

If your reason for wanting a timing rule is that deep falls frighten you, that is a real problem deserving a real answer. But the answer is more likely to be owning an amount of equity you can live with through a 59.5% fall than a rule trading 7.5 times a year to soften it. You will be tempted to abandon that rule at exactly the moment it matters, and a rule abandoned halfway charges you everything and returns nothing.

Disclaimer

Educational content only. This is not personalised financial, investment or tax advice. Figures quoted are historical or illustrative and are not forecasts. Consult a qualified professional before acting on anything you read here.