SCE Reversals

This tool uses past market data to attempt to identify where changes in “memory” may occur to spot reversals. The Hurst Exponent was a big inspiration for this code. The main driver is identifying when past ranges expand and contract, leading to a change in direction. With the use of Sum of Squared Errors, users do not need to input anything.

Getting optimized parameters

// Define ranges for N and lkb
N_range = array.from(15, 20, 25, 30, 35, 40, 45, 50, 55, 60)

// Function to calculate SSE
sse_calc(_N) =>
    x = math.pow(close - close[_N], 2)
    y = math.pow(close - close[_N], 2) + math.pow(close, 2)
    z = x / y
    scaled_z = z * math.log(_N)
    min_r = ta.lowest(scaled_z, _N)
    max_r = ta.highest(scaled_z, _N)
    norm_r = (scaled_z - min_r) / (max_r - min_r)

    SMA = ta.sma(close, _N)
    reversal_bullish = norm_r[1] == 1.000 and norm_r < 0.90 and close < SMA and session.ismarket and barstate.isconfirmed
    reversal_bearish = norm_r[1] == 1.000 and norm_r < 0.90 and close > SMA and session.ismarket and barstate.isconfirmed

    var float error = na
    if reversal_bullish or reversal_bearish
        error := math.pow(close - SMA, 2)
        error

    else
        error := 999999999999999999999999999999999999999
        error

    error

var int N_opt = na
var float min_SSE = na
// Loop through ranges and calculate SSE
for N in N_range

    sse = sse_calc(N)
    if na(min_SSE) or sse < min_SSE
        min_SSE := sse
        N_opt := N

The N_range list encompasses every lookback value to check with. The sse_calc function accepts an individual element to then perform the calculation for Reversals. If there is a reversal, the error becomes how far away the close is from a moving average with that look back. Lowest error wins. That would be the look back used for the Reversals calculation.

Reversals calculation

// Calculating with optimized parameters
x_opt = math.pow(close - close[N_opt], 2)
y_opt = math.pow(close - close[N_opt], 2) + math.pow(close, 2)
z_opt = x_opt / y_opt
scaled_z_opt = z_opt * math.log(N_opt)
min_r_opt = ta.lowest(scaled_z_opt, N_opt)
max_r_opt = ta.highest(scaled_z_opt, N_opt)
norm_r_opt = (scaled_z_opt - min_r_opt) / (max_r_opt - min_r_opt)

SMA_opt = ta.sma(close, N_opt)

reversal_bullish_opt = norm_r_opt[1] == 1.000 and norm_r_opt < 0.90 and close < SMA_opt and close > high[1] and close > open and session.ismarket and barstate.isconfirmed
reversal_bearish_opt = norm_r_opt[1] == 1.000 and norm_r_opt < 0.90 and close > SMA_opt and close < low[1] and close < open and session.ismarket and barstate.isconfirmed

X_opt and y_opt are the compared values to develop the system. Everything done afterwards is scaling and using it to spot the Reversals. X_opt is the current close, minus the close with the optimal N bars back, squared. Then y_opt is also that but plus the current close squared. Z_opt is then x_opt / y_opt. This gives us a pretty small number that will go up when we approach tops or bottoms. To make life a little easier I normalize the value between 0 and 1.

After I find the moving average with the optimal N, I can check if there is a Reversal. Reversals are there when the last value is at 1 and the current value drops below 0.90. This would tell us that “memory” was strong and is now changing. To determine direction and help with accuracy, if the close is above the moving average it is a bearish alert, and vice versa. As well as the close must be below the last low for a bearish Reversal, above the last high for a bullish Reversal. Also the close must be above the open for a bullish Reversal, and below for a bearish one.

Visual examples



This TSLA chart shows how alerts may come around. The bullish and bearish labels are plotted on the chart along with a reference line to see price interact with.



The indicator has the potential to be inactive, like we see here on $OKLO. There is only one alert, and it marks the bottom nicely.



Stocks with strong trends like NOW may be more susceptible to false alerts. Assets that are volatile and bounce around a lot may be better.



It works on intra day charts the same as on Daily or longer charts. We see here on QQQ it spotted the bottom on this particular trading day.

This tool is meant to aid traders in making decisions, not to be followed blindly. No trading tool is 100% accurate and Sum of Squared Errors does not guarantee the most optimal value. I encourage feedback and constructive criticism.