Stable Correlations

stable_correlations.py

meansize = 4
corrsize = 4
mediansize = 4

commodities_rolling_mean_df = commodities_df.rolling(meansize).mean().dropna(thresh = len(commodities_df)-meansize-corrsize+2)
commodities_rolling_roll_corr_df = commodities_rolling_mean_df.rolling(corrsize).corr()
commodities_rolling_roll_corr_df

correlations = pd.DataFrame()
#correlations_pre = pd.DataFrame()
for c in commodities_rolling_roll_corr_df.columns:
    #print(c)
    aggregated = pd.DataFrame()
    for i in commodities_rolling_mean_df.tail(len(commodities_df)-meansize-corrsize+2).index:
        #print(i,c)
        subset = commodities_rolling_roll_corr_df.loc[i,c]
        
        aggregated = pd.concat([pd.DataFrame(subset),aggregated],axis=1)
            
    median_cropped = aggregated.T[::-1].rolling(mediansize).median().tail(-corrsize-1)
    #correlations_pre = pd.concat([correlations_pre,median_cropped],axis=1)
    size_median_cropped_n = len(median_cropped)
    
    correlations = pd.concat([correlations,median_cropped.median(axis=0)],axis=1)

correlations.columns = commodities_rolling_roll_corr_df.columns

crit_one_way = critical_95t2(mediansize)[0]
crit_two_way = critical_95t2(mediansize)[1]

highlights = (np.round(abs(correlations),2)!=1)>crit_one_way

correlations_masked = correlations.copy()
for c in correlations.columns:
    for i in correlations.index:
        #print(i,c)
        if highlights.loc[i,c] == True:
            correlations_masked.loc[i,c] = correlations_masked.loc[i,c]
        else:
            correlations_masked.loc[i,c] = ""

#flow_path = find_max_flow_pairs(correlations_masked)
correlations_masked