Rolling Window of Local Minima/Maxima

Question

I've made a script (shown below) that helps determine local maxima points using historical stock data. It uses the daily highs to mark out local resistance levels. Works great, but what I would like is, for any given point in time (or row in the stock data), I want to know what the most recent resistance level was just prior to that point. I want this in it's own column in the dataset. So for instance:

The top grey line is the highs for each day, and the bottom grey line was the close of each day. So roughly speaking, the dataset for that section would look like this:

High            Close
216.8099976     216.3399963
215.1499939     213.2299957
214.6999969     213.1499939
215.7299957     215.2799988 <- First blue dot at high
213.6900024     213.3699951
214.8800049     213.4100037 <- 2nd blue dot at high
214.5899963     213.4199982 
216.0299988     215.8200073
217.5299988     217.1799927 <- 3rd blue dot at high
216.8800049     215.9900055
215.2299957     214.2400055
215.6799927     215.5700073
....

Right now, this script looks at the entire dataset at once to determine the local maxima indexes for the highs, and then for any given point in the stock history (ie any given row), it looks for the NEXT maxima in the list of all maximas found. This would be a way to determine where the next resistance level is, but I don't want that due to a look ahead bias. I just want to have a column of the most recent past resistance level or maybe even the latest 2 recent points in 2 columns. That would be ideal actually.

So my final output would look like this for the 1 column:

High            Close           Most_Rec_Max
216.8099976     216.3399963     0
215.1499939     213.2299957     0
214.6999969     213.1499939     0
215.7299957     215.2799988     0
213.6900024     213.3699951     215.7299957
214.8800049     213.4100037     215.7299957
214.5899963     213.4199982     214.8800049
216.0299988     215.8200073     214.8800049
217.5299988     217.1799927     214.8800049
216.8800049     215.9900055     217.5299988
215.2299957     214.2400055     217.5299988
215.6799927     215.5700073     217.5299988
....

You'll notice that the dot only shows up in most recent column after it has already been discovered.

Here is the code I am using:

real_close_prices = df['Close'].to_numpy()

highs = df['High'].to_numpy()

max_indexes = (np.diff(np.sign(np.diff(highs))) < 0).nonzero()[0] + 1 # local max
# +1 due to the fact that diff reduces the original index number

max_values_at_indexes = highs[max_indexes]
curr_high = [c for c in highs]
max_values_at_indexes.sort()
for m in max_values_at_indexes:
    for i, c in enumerate(highs):
        if m > c and curr_high[i] == c:
            curr_high[i] = m
#print(nextbig)
df['High_Resistance'] = curr_high

# plot
plt.figure(figsize=(12, 5))
plt.plot(x, highs, color='grey')
plt.plot(x, real_close_prices, color='grey')
plt.plot(x[max_indexes], highs[max_indexes], "o", label="max", color='b')
plt.show()

Hoping someone will be able to help me out with this. Thanks!

Answer 1

Here is one approach. Once you know where the peaks are, you can store peak indices in p_ids and peak values in p_vals . To assign the k 'th most recent peak, note that p_vals[:-k] will occur at p_ids[k:] . The rest is forward filling.

# find all local maxima in the series by comparing to shifted values
peaks = (df.High > df.High.shift(1)) & (df.High > df.High.shift(-1))
# pass peak value if peak is achieved and NaN otherwise
# forward fill with previous peak value & handle leading NaNs with fillna
df['Most_Rec_Max'] = (df.High * peaks.replace(False, np.nan)).ffill().fillna(0)

# for finding n-most recent peak
p_ids, = np.where(peaks)
p_vals = df.High[p_ids].values
for n in [1,2]:
  col_name = f'{n+1}_Most_Rec_Max'
  df[col_name] = np.nan
  df.loc[p_ids[n:], col_name] = p_vals[:-n]
  df[col_name].ffill(inplace=True)
  df[col_name].fillna(0, inplace=True)


#           High       Close  Most_Rec_Max  2_Most_Rec_Max  3_Most_Rec_Max
# 0   216.809998  216.339996      0.000000        0.000000        0.000000
# 1   215.149994  213.229996      0.000000        0.000000        0.000000
# 2   214.699997  213.149994      0.000000        0.000000        0.000000
# 3   215.729996  215.279999    215.729996        0.000000        0.000000
# 4   213.690002  213.369995    215.729996        0.000000        0.000000
# 5   214.880005  213.410004    214.880005      215.729996        0.000000
# 6   214.589996  213.419998    214.880005      215.729996        0.000000
# 7   216.029999  215.820007    214.880005      215.729996        0.000000
# 8   217.529999  217.179993    217.529999      214.880005      215.729996
# 9   216.880005  215.990006    217.529999      214.880005      215.729996
# 10  215.229996  214.240006    217.529999      214.880005      215.729996
# 11  215.679993  215.570007    217.529999      214.880005      215.729996

Answer 2

I just came across this function that might help you a lot: scipy.signal.find_peaks .

Based on your sample dataframe, we can do the following:

from scipy.signal import find_peaks

## Grab the minimum high value as a threshold.
min_high = df["High"].min()

### Run the High values through the function. The docs explain more,
### but we can set our height to the minimum high value.
### We just need one out of two return values.

peaks, _ = find_peaks(df["High"], height=min_high)

### Do some maintenance and add a column to mark peaks

# Merge on our index values
df1 = df.merge(peaks_df, how="left", left_index=True, right_index=True)

# Set non-null values to 1 and null values to 0; Convert column to integer type.
df1.loc[~df1["local_high"].isna(), "local_high"] = 1
df1.loc[df1["local_high"].isna(), "local_high"] = 0
df1["local_high"] = df1["local_high"].astype(int)

Then, your dataframe should look like the following:

          High         Low  local_high
0   216.809998  216.339996           0
1   215.149994  213.229996           0
2   214.699997  213.149994           0
3   215.729996  215.279999           1
4   213.690002  213.369995           0
5   214.880005  213.410004           1
6   214.589996  213.419998           0
7   216.029999  215.820007           0
8   217.529999  217.179993           1
9   216.880005  215.990005           0
10  215.229996  214.240005           0
11  215.679993  215.570007           0