forex-scalping.com

RSI 70 means average gains are 2.33 times average losses: the ratio in pips on M5

RSI looks like a mysterious 0 to 100 number, but it's a ratio in disguise. RSI = 100 x RS / (1 + RS), where RS is the average gain divided by the average loss over the period, 14 bars by default, with Wilder's smoothing.

Turn it round and you get RS = RSI / (100 minus RSI):

RSI 30: RS = 0.43, so average gains are 43% of average losses.
RSI 50: RS = 1.
RSI 70: RS = 2.33.
RSI 80: RS = 4.
RSI 90: RS = 9.

An example on M5: the average up-move over the last 14 bars is 2.8 pips and the average down-move is 1.2 pips. RS is 2.8 / 1.2 = 2.33, and RSI is 70. If the numbers were 0.7 pips and 0.3 pips, RSI would still be 70. The indicator has no idea how large the moves are, only their proportion.

That matters for scalping, because RSI 70 in a dead hour with 0.7 pip average gains is much less informative than RSI 70 with 2.8 pip gains. One is a ratio of tiny moves that a 0.8 pip spread swallows. A simple filter is to read the ATR next to the RSI and only use overbought and oversold readings when the average bar is bigger than a couple of spreads.

Going from 70 to 80 needs the ratio to go from 2.33 to 4, a much bigger change than going from 60 to 70, where it moves from 1.5 to 2.33. The scale is compressed at the ends.

The ratio view explains why RSI can stay at 70 or above in a strong trend. If average gains remain at 2.33 times average losses for a long time, the reading stays high without being a sell signal. To push RSI from 70 back to 50, the average gain must fall to equal the average loss, which takes a number of bars of the same size up and down. In a trend with steady 2:1 up bars, the RSI just sits. So the oscillator tells you about the balance of moves, and doesn't know about a limit.

Smoothing changes how fast it reacts. Wilder's version updates each average as (previous average x 13 + new value) / 14, so one large bar has a limited effect. Say the average gain is 2.8 and the average loss is 1.2, RSI 70, and the next bar is an up bar of 6 pips. The average gain becomes (13 x 2.8 + 6) / 14 = 3.03 and the average loss becomes 13 x 1.2 / 14 = 1.11. RS moves from 2.33 to 2.72 and RSI from 70 to 73.1. A single big candle moves the reading by about three points, and that damping is a deliberate feature.

A way to use the idea is to build your own version in a spreadsheet for 14 closes. Compute up moves and down moves, average them, take the ratio, and compare with your platform's RSI value. They won't match exactly until you apply the Wilder smoothing over a longer history, but within a point or two they show that you understand the formula. Doing it once makes the numbers on the screen feel less arbitrary, and you can set thresholds by the ratio you want, for instance gains at three times losses, which is RSI 75.

Because RSI depends on the period, 14 bars on M5 is 70 minutes, while 14 bars on M1 is 14 minutes. The same ratio gives the same number, but the window is different, so the readings aren't comparable across charts. A 7 period RSI on M5 looks at 35 minutes and swings between extremes more often. If you change the period, check how many times a day RSI crosses 70 or 30 with each setting in your last week of data, because the count of signals changes a lot, and so does the reliability you can expect.