Why does changing 14 to 20 change the whole picture?

RSI(14). MACD(12, 26, 9). The 50-day and the 200-day. These numbers are printed in every textbook with the authority of physical constants. They are conventions — defaults that survived because people repeated them.

Where the famous numbers came from

• 14 comes from Welles Wilder's 1978 book, which introduced RSI, ATR and ADX. He chose it from his own work on 1970s commodity data, before cheap computing existed. It was a judgement, not a derivation.
• 12, 26 and 9 for MACD date from the late 1970s and are usually explained as roughly two weeks and one month back when exchanges traded six-day weeks. The story may be folklore; either way, nobody claims the values fell out of a law of markets.
• 50 and 200 are round numbers that land near ten trading weeks and near ten months of trading days. (Fifty is not a quarter: a calendar quarter holds about 63 trading days.) Their power today comes substantially from how many people watch them.

What the setting actually changes

Consider a stock in a long uptrend that has a sharp four-day pullback — the last one on the schematic below. Illustratively, RSI(9) might drop to about 35 while RSI(21), over the same bars, barely dips to about 50. One chart says "sharp weakness", the other says "nothing happened".

Neither is wrong. They answer different questions: fast by nine-day standards versus fast by twenty-one-day standards. The parameter is the question. Choosing it silently means asking a question you never articulated, then treating the answer as objective.

The curve-fitting trap

The tempting fix is to test them all. Sweep the fast moving average from 5 to 100 and the slow one from 20 to 300 and you have tens of thousands of combinations. One of them will look magnificent on your history. One of them looks magnificent on random data too.

Do the arithmetic of luck. Test 200 variants where each has a 5% chance of looking impressive purely by chance, and you should expect about 10 impressive results from nothing at all. The winner isn't a discovery — it's a selection. And the selection process is invisible in the final chart, which shows only the survivor.

Three tells of a curve-fitted parameter

1. It's a spike, not a plateau. 14 is wonderful, 13 and 15 are terrible. Real effects are rarely that fussy; noise always is.
2. It moves when the window moves. Shift the test period six months and the "optimal" value changes. An edge that depends on the start date is a description of one sample.
3. It arrived after many uncounted attempts. If you can't say how many combinations you looked at, you can't interpret the one you kept.

The stance that survives contact

Prefer plateaus over spikes. Keep the defaults unless you can say what question changed. Test on data you didn't use while searching. And hold the ceiling firmly in mind: no parameter search converts a past pattern into a promise. The Risk Management for Traders course takes this apart properly under the heading backtest ≠ future; here it is enough to know that the knobs on your indicators are choices you are responsible for.

In the data

Three averages of the same Apple closes, windows of 20, 50 and 100 sessions:

Same prices, same formula, three different verdicts on where "the average" is, and they usually sit furthest apart after a sharp move. A column of indicator values carries no record of the window that produced it, so write the setting beside every value you keep. Do not assume a tool's default is the textbook number either: some default every window to 50, and then the famous 14 has to be asked for.

Try it now

1. Find the four-bar pullback on the schematic above and decide, from the bars alone, whether it is "sharp weakness" or "nothing happened". You cannot — the bars do not contain that word. The window you would have chosen contains it.
2. Now make the setting move the answer on prices nobody arranged. Pick a sharp pullback in the year below, where the pane under the bars draws RSI at 7, 14 and 21 sessions on one 0–100 scale, and find one date where the three disagree about whether the move was extreme.

3. Write one sentence naming the question your chosen setting answers. If you can't, you chose it because it was top of the menu — which is a fine reason to start and a bad reason to stay.