Technical analysis of stock charts is based on the hypothesis of the existence of repeating patterns, patterns that can bring profit. We saw two black candles and three white candles in a certain order - open and get money. But for some reason there is little information (or it is absent altogether) about the proof of this hypothesis and how and on the basis of what it was generally formed. Those. how many times on historical data this or that figure has worked as it should. Where and who decided that there is a falling triangle, tunnels, a head with shoulders, etc. Who checked their profitability on the quotes accumulated over the entire existence of the exchange? Why not two parallel lines and a parallelepiped or a circle with an oval? We will talk about this proof in this article, i.e. run the python script to find patterns,we obtain historical data by copying charts from the room.
Let's start with one of the most popular patterns - head and shoulders on 6-hour charts for 10 currency pairs.
It is necessary to bring the statistics to the form of DataFrame in the format (this is an excerpt from the end, data on 10 currency pairs at 6 hours on average since 2005 will take about 160 thousand lines)
|
|
TIME |
OPEN |
HIGH |
LOW |
CLOSE |
TITLE |
|---|---|---|---|---|---|
16/09/19 06:00:00 |
1.3213 |
1.3234 |
1.3211 |
1.3232 |
USD_CAD_6H |
16/09/19 12:00:00 |
1.3230 |
1.3265 |
1.3228 |
1.3257 |
USD_CAD_6H |
16/09/19 18:00:00 |
1.3255 |
1.3270 |
1.3231 |
1.3247 |
USD_CAD_6H |
17/09/19 00:00:00 |
1.3248 |
1.3249 |
1.3232 |
1.3239 |
USD_CAD_6H |
17/09/19 06:00:00 |
1.3238 |
1.3253 |
1.3237 |
1.3247 |
USD_CAD_6H |
17/09/19 12:00:00 |
1.3249 |
1.3259 |
1.3240 |
1.3255 |
USD_CAD_6H |
17/09/19 18:00:00 |
1.3256 |
1.3298 |
1.3231 |
1.3247 |
USD_CAD_6H |
18/09/19 00:00:00 |
1.3245 |
1.3252 |
1.3237 |
1.3239 |
USD_CAD_6H |
18/09/19 06:00:00 |
1.3240 |
1.3258 |
1.3238 |
1.3254 |
USD_CAD_6H |
18/09/19 12:00:00 |
1.3252 |
1.3270 |
1.3249 |
1.3250 |
USD_CAD_6H |
18/09/19 18:00:00 |
1.3252 |
1.3285 |
1.3238 |
1.3273 |
USD_CAD_6H |
19/09/19 00:00:00 |
1.3277 |
1.3308 |
1.3267 |
1.3295 |
USD_CAD_6H |
19/09/19 06:00:00 |
1.3297 |
1.3305 |
1.3279 |
1.3282 |
USD_CAD_6H |
19/09/19 12:00:00 |
1.3281 |
1.3292 |
1.3272 |
1.3273 |
USD_CAD_6H |
19/09/19 18:00:00 |
1.3274 |
1.3279 |
1.3238 |
1.3243 |
USD_CAD_6H |
20/09/19 00:00:00 |
1.3244 |
1.3268 |
1.3242 |
1.3257 |
USD_CAD_6H |
20/09/19 06:00:00 |
1.3259 |
1.3271 |
1.3256 |
1.3266 |
USD_CAD_6H |
20/09/19 12:00:00 |
1.3265 |
1.3275 |
1.3256 |
1.3264 |
USD_CAD_6H |
20/09/19 18:00:00 |
1.3262 |
1.3299 |
1.3254 |
1.3255 |
USD_CAD_6H |
21/09/19 00:00:00 |
1.3253 |
1.3278 |
1.3251 |
1.3261 |
USD_CAD_6H |
. , , .. , , .. , , , , , .. , . , , , , - … , "" , .. . . , , ( ). , , .. , , - H&S ( ).
#,
list_H_AND_S=[]
# ( , , )
list_NOT_H_AND_S=[]
for i in range(600,len(df_ALL['CLOSE'])-600):
# , ..
if max(df_ALL['HIGH'][i-30:i+15])==df_ALL['HIGH'][i]:
# 15,
for z in range (15,200,15):
left_shoulder=max(df_ALL['HIGH'][i-30:i+15])
#
left_shoulder_index=df_ALL['HIGH'][i-30:i+15].idxmax()
# ,
if max(df_ALL['HIGH'][df_ALL['LOW'][left_shoulder_index:left_shoulder_index+z+1].idxmin():left_shoulder_index+z+1])>left_shoulder and\
(left_shoulder-min(df_ALL['LOW'][left_shoulder_index:left_shoulder_index+z+1]))/((max(df_ALL['HIGH'][df_ALL['LOW'][left_shoulder_index:left_shoulder_index+z+1].idxmin():left_shoulder_index+z+1])-min(df_ALL['LOW'][left_shoulder_index:left_shoulder_index+z+1])))>=0.6:
head=max(df_ALL['HIGH'][df_ALL['LOW'][left_shoulder_index:left_shoulder_index+z+1].idxmin():left_shoulder_index+z+1])
#
head_index=df_ALL['HIGH'][df_ALL['LOW'][left_shoulder_index:left_shoulder_index+z+1].idxmin():left_shoulder_index+z+1].idxmax()
for b in range (15,200,15):
# , ..
first_bottom= min(df_ALL['LOW'][left_shoulder_index:head_index+1])
# 1-
first_bottom_index=df_ALL['LOW'][left_shoulder_index:head_index+1].idxmin()
#
if min(df_ALL['LOW'][head_index:head_index+b])<first_bottom and \
(head-first_bottom)/(head-min(df_ALL['LOW'][head_index:head_index+b]))>=0.5:
second_bottom= min(df_ALL['LOW'][head_index:head_index+b])
# 2-
second_bottom_index=df_ALL['LOW'][head_index:head_index+b].idxmin()
for o in range(2,300):
# max(df_ALL['HIGH'][second_bottom_index:second_bottom_index+o+1]) -
if max(df_ALL['HIGH'][second_bottom_index:second_bottom_index+o+1])>=left_shoulder and \
max(df_ALL['HIGH'][second_bottom_index:second_bottom_index+o+1])<head and \
second_bottom_index+o-df_ALL['HIGH'][second_bottom_index:second_bottom_index+o+1].idxmax()>0 and \
(second_bottom_index+o+1)-df_ALL['LOW'][df_ALL['HIGH'][second_bottom_index:second_bottom_index+o+1].idxmax()]>2 and \
min(df_ALL['LOW'][df_ALL['HIGH'][second_bottom_index:second_bottom_index+o+1].idxmax():second_bottom_index+o+1])==df_ALL['LOW'][second_bottom_index+o] and \
min(df_ALL['LOW'][df_ALL['HIGH'][second_bottom_index:second_bottom_index+o+1].idxmax():second_bottom_index+o])>second_bottom and \
df_ALL['CLOSE'][second_bottom_index+o]<second_bottom and \
min(df_ALL['LOW'][second_bottom_index:df_ALL['HIGH'][second_bottom_index:second_bottom_index+o+1].idxmax()+1])==second_bottom and \
min(df_ALL['LOW'][left_shoulder_index:head_index])==first_bottom and \
max(df_ALL['HIGH'][first_bottom_index:second_bottom_index+1])==head and \
df_ALL['CLOSE'][second_bottom_index+o]-df_ALL['OPEN'][second_bottom_index+o]<=0:
# , H$S
for x in range (2,300):
# ,
# , (- )*2. ,
# -
if max(df_ALL['HIGH'][second_bottom_index+o:second_bottom_index+o+x+1])<=max(df_ALL['HIGH'][second_bottom_index:second_bottom_index+o+1]) and \
min(df_ALL['LOW'][second_bottom_index+o:second_bottom_index+o+x+1])<=second_bottom-(head-second_bottom)*2:
if (df_ALL['TIME'][left_shoulder_index]+' '+ df_ALL['TIME'][head_index]+' '+df_ALL['TIME'][second_bottom_index]+' '+df_ALL['TIME'][second_bottom_index+o]+' '+df_ALL['TITLE'][i]+' '+df_ALL['TITLE'][second_bottom_index+o]+' '+str(df_ALL['CLOSE'][o+second_bottom_index]-df_ALL['OPEN'][o+second_bottom_index])) not in list_H_AND_S:
list_H_AND_S.append(df_ALL['TIME'][left_shoulder_index]+' '+ df_ALL['TIME'][head_index]+' '+df_ALL['TIME'][second_bottom_index]+' '+df_ALL['TIME'][second_bottom_index+o]+' '+df_ALL['TITLE'][i]+' '+df_ALL['TITLE'][second_bottom_index+o]+' '+str(df_ALL['CLOSE'][o+second_bottom_index]-df_ALL['OPEN'][o+second_bottom_index]))
elif max(df_ALL['HIGH'][second_bottom_index+o:second_bottom_index+o+x+1])>max(df_ALL['HIGH'][second_bottom_index:second_bottom_index+o+1]) and \
min(df_ALL['LOW'][second_bottom_index+o:second_bottom_index+o+x+1])>first_bottom-(head-second_bottom):
if (df_ALL['TIME'][left_shoulder_index]+' '+ df_ALL['TIME'][head_index]+' '+df_ALL['TIME'][second_bottom_index]+' '+df_ALL['TIME'][second_bottom_index+o]+' '+df_ALL['TITLE'][i]+' '+df_ALL['TITLE'][second_bottom_index+o]+' '+str(df_ALL['CLOSE'][o+second_bottom_index]-df_ALL['OPEN'][o+second_bottom_index])) not in list_NOT_H_AND_S and \
(df_ALL['TIME'][left_shoulder_index]+' '+ df_ALL['TIME'][head_index]+' '+df_ALL['TIME'][second_bottom_index]+' '+df_ALL['TIME'][second_bottom_index+o]+' '+df_ALL['TITLE'][i]+' '+df_ALL['TITLE'][second_bottom_index+o]+' '+str(df_ALL['CLOSE'][o+second_bottom_index]-df_ALL['OPEN'][o+second_bottom_index])) not in list_H_AND_S:
list_NOT_H_AND_S.append(df_ALL['TIME'][left_shoulder_index]+' '+ df_ALL['TIME'][head_index]+' '+df_ALL['TIME'][second_bottom_index]+' '+df_ALL['TIME'][second_bottom_index+o]+' '+df_ALL['TITLE'][i]+' '+df_ALL['TITLE'][second_bottom_index+o]+' '+str(df_ALL['CLOSE'][o+second_bottom_index]-df_ALL['OPEN'][o+second_bottom_index]))
len(list_H_AND_S),len(list_NOT_H_AND_S)
1/6 ( ), ( -) , . . , , , - H&S , …
(/)
, , . 1 10, . ,
#
list_vimpel=[]
#
list_NOT_vimpel=[]
for i in range(600,len(df_ALL['CLOSE'])-1000):
if max(df_ALL['HIGH'][i-15:i+15])==df_ALL['HIGH'][i]:
for z in range (15,300,15):
#
first_max=max(df_ALL['HIGH'][i-15:i+15])
# 1-
first_max_index=df_ALL['HIGH'][i-15:i+15].idxmax()
if (first_max-min(df_ALL['LOW'][first_max_index:first_max_index+z+1]))>0 and \
max(df_ALL['HIGH'][df_ALL['LOW'][first_max_index:first_max_index+z+1].idxmin():first_max_index+z+1])>first_max and \
(max(df_ALL['HIGH'][df_ALL['LOW'][first_max_index:first_max_index+z+1].idxmin():first_max_index+z+1])-min(df_ALL['LOW'][first_max_index:first_max_index+z+1]))/(first_max-min(df_ALL['LOW'][first_max_index:first_max_index+z+1]))<=2.2 and \
(max(df_ALL['HIGH'][df_ALL['LOW'][first_max_index:first_max_index+z+1].idxmin():first_max_index+z+1])-min(df_ALL['LOW'][first_max_index:first_max_index+z+1]))/(first_max-min(df_ALL['LOW'][first_max_index:first_max_index+z+1]))>=1.3:
second_max=max(df_ALL['HIGH'][df_ALL['LOW'][first_max_index:first_max_index+z+1].idxmin():first_max_index+z+1])
second_max_index=df_ALL['HIGH'][df_ALL['LOW'][first_max_index:first_max_index+z+1].idxmin():first_max_index+z+1].idxmax()
# ,
first_min= min(df_ALL['LOW'][first_max_index:second_max_index+1])
# 1-
first_min_index=df_ALL['LOW'][first_max_index:second_max_index+1].idxmin()
for o in range(2,300):
if (second_max_index-first_min_index)>=25 and \
min(df_ALL['LOW'][second_max_index:second_max_index+o+1])==df_ALL['LOW'][o+second_max_index] and \
df_ALL['LOW'][o+second_max_index]>first_min and \
df_ALL['HIGH'][o+second_max_index]<second_max and \
(second_max_index-first_min_index)>=25 and \
max(df_ALL['HIGH'][first_min_index:o+second_max_index+1])==second_max and \
min(df_ALL['LOW'][first_min_index:second_max_index+1])==df_ALL['LOW'][first_min_index] and \
max(df_ALL['HIGH'][first_max_index:first_min_index+1])==first_max and \
(o+second_max_index)-second_max_index>=16 and \
df_ALL['LOW'][o+second_max_index]<=(first_min+((second_max-first_max)*(((o+second_max_index)-first_min_index)/(second_max_index-first_max_index)))) and \
df_ALL['CLOSE'][o+second_max_index]>=(first_min+((second_max-first_max)*(((o+second_max_index)-first_min_index)/(second_max_index-first_max_index)))):
for x in range (2,300):
if min(df_ALL['LOW'][o+second_max_index:o+second_max_index+x+1])==df_ALL['LOW'][o+second_max_index] and \
max(df_ALL['HIGH'][o+second_max_index:o+second_max_index+x+1])>=second_max:
if (df_ALL['TIME'][first_max_index]+' '+ df_ALL['TIME'][second_max_index]+' '+df_ALL['TIME'][second_max_index+o]+' '+df_ALL['TITLE'][i]+' '+df_ALL['TITLE'][second_max_index+o]+' '+str(df_ALL['CLOSE'][o+second_max_index]-df_ALL['OPEN'][o+second_max_index])+' '+str((second_max-first_min)/(first_max-first_min))+' '+str((second_max_index-first_min_index)/(first_min_index-first_max_index))+' '+str(((second_max-first_min)/(first_max-first_min))/((second_max-df_ALL['LOW'][o+second_max_index])/(second_max-first_min)))+' '+str(df_ALL['WeekDay'][o+second_max_index])+' '+str((second_max_index-first_max_index)/((o+second_max_index)-second_max_index))) not in list_vimpel:
list_vimpel.append(df_ALL['TIME'][first_max_index]+' '+ df_ALL['TIME'][second_max_index]+' '+df_ALL['TIME'][second_max_index+o]+' '+df_ALL['TITLE'][i]+' '+df_ALL['TITLE'][second_max_index+o]+' '+str(df_ALL['CLOSE'][o+second_max_index]-df_ALL['OPEN'][o+second_max_index])+' '+str((second_max-first_min)/(first_max-first_min))+' '+str((second_max_index-first_min_index)/(first_min_index-first_max_index))+' '+str(((second_max-first_min)/(first_max-first_min))/((second_max-df_ALL['LOW'][o+second_max_index])/(second_max-first_min)))+' '+str(df_ALL['WeekDay'][o+second_max_index])+' '+str((second_max_index-first_max_index)/((o+second_max_index)-second_max_index)))
elif min(df_ALL['LOW'][o+second_max_index:o+second_max_index+x+1])<df_ALL['LOW'][o+second_max_index] and \
max(df_ALL['HIGH'][o+second_max_index:o+second_max_index+x+1])<second_max:
if (df_ALL['TIME'][first_max_index]+' '+ df_ALL['TIME'][second_max_index]+' '+df_ALL['TIME'][second_max_index+o]+' '+df_ALL['TITLE'][i]+' '+df_ALL['TITLE'][second_max_index+o]+' '+str(df_ALL['CLOSE'][o+second_max_index]-df_ALL['OPEN'][o+second_max_index])+' '+str((second_max-first_min)/(first_max-first_min))+' '+str((second_max_index-first_min_index)/(first_min_index-first_max_index))+' '+str(((second_max-first_min)/(first_max-first_min))/((second_max-df_ALL['LOW'][o+second_max_index])/(second_max-first_min)))+' '+str(df_ALL['WeekDay'][o+second_max_index])+' '+str((second_max_index-first_max_index)/((o+second_max_index)-second_max_index))) not in list_NOT_vimpel and \
(df_ALL['TIME'][first_max_index]+' '+ df_ALL['TIME'][second_max_index]+' '+df_ALL['TIME'][second_max_index+o]+' '+df_ALL['TITLE'][i]+' '+df_ALL['TITLE'][second_max_index+o]+' '+str(df_ALL['CLOSE'][o+second_max_index]-df_ALL['OPEN'][o+second_max_index])+' '+str((second_max-first_min)/(first_max-first_min))+' '+str((second_max_index-first_min_index)/(first_min_index-first_max_index))+' '+str(((second_max-first_min)/(first_max-first_min))/((second_max-df_ALL['LOW'][o+second_max_index])/(second_max-first_min)))+' '+str(df_ALL['WeekDay'][o+second_max_index])+' '+str((second_max_index-first_max_index)/((o+second_max_index)-second_max_index))) not in list_vimpel:
list_NOT_vimpel.append(df_ALL['TIME'][first_max_index]+' '+ df_ALL['TIME'][second_max_index]+' '+df_ALL['TIME'][second_max_index+o]+' '+df_ALL['TITLE'][i]+' '+df_ALL['TITLE'][second_max_index+o]+' '+str(df_ALL['CLOSE'][o+second_max_index]-df_ALL['OPEN'][o+second_max_index])+' '+str((second_max-first_min)/(first_max-first_min))+' '+str((second_max_index-first_min_index)/(first_min_index-first_max_index))+' '+str(((second_max-first_min)/(first_max-first_min))/((second_max-df_ALL['LOW'][o+second_max_index])/(second_max-first_min)))+' '+str(df_ALL['WeekDay'][o+second_max_index])+' '+str((second_max_index-first_max_index)/((o+second_max_index)-second_max_index)))
len(list_vimpel),len(list_NOT_vimpel)
The purpose of this article is not to reduce technical analysis to pseudo-theories, but rather to show the unscientific nature of many articles and textbooks that claim that popular patterns work due to the lack of an answer to the main question - why does this combination of candles work? Where did this information come from? Based on personal experience, how is the belief in a creator proven in Scientology as a science? Perhaps someone discovered (not without proof, of course) a pattern that brings at least some profit, as they say, write in the comments. The opinion of representatives of the exchange sphere is especially welcome. Thanks for attention!
PS The author understands that the code may not be written perfectly, not optimized, etc.