Chuanshuoge

Tuesday, 10 December 2019

Sunday, 8 December 2019

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt

data = get_pricing('SPY', start_date='2015-6-1', end_date='2015-7-1', fields='volume')
x = np.arange(len(data))
print data.values
print data.index
plt.bar(x, data.values)

tick_num = 12
tick_step = float(len(data))/tick_num
x_tick = np.arange(0, len(data), tick_step)
x_label = []
for i in range(0, tick_num):
x_label.append(data.index[int(x_tick[i])].date())

print x_tick
print x_label

plt.xticks(x_tick, x_label)
plt.xticks(rotation=-90)
plt.ylabel('Volume (USD)');

[ 7.63173340e+07 7.56557330e+07 7.05195340e+07 1.07026621e+08
8.29844740e+07 7.15034260e+07 7.82032850e+07 1.01090613e+08
5.82412800e+07 8.38160790e+07 9.04484390e+07 6.10451090e+07
9.68857910e+07 1.11951254e+08 7.89226310e+07 5.74651720e+07
4.89672890e+07 6.57530870e+07 6.33106560e+07 7.72355920e+07
1.41736435e+08 1.27227662e+08 9.46635640e+07]
DatetimeIndex(['2015-06-01', '2015-06-02', '2015-06-03', '2015-06-04',
'2015-06-05', '2015-06-08', '2015-06-09', '2015-06-10',
'2015-06-11', '2015-06-12', '2015-06-15', '2015-06-16',
'2015-06-17', '2015-06-18', '2015-06-19', '2015-06-22',
'2015-06-23', '2015-06-24', '2015-06-25', '2015-06-26',
'2015-06-29', '2015-06-30', '2015-07-01'],
dtype='datetime64[ns, UTC]', freq='C')
[ 0. 1.91666667 3.83333333 5.75 7.66666667
9.58333333 11.5 13.41666667 15.33333333 17.25 19.16666667
21.08333333]
[datetime.date(2015, 6, 1), datetime.date(2015, 6, 2), datetime.date(2015, 6, 4), datetime.date(2015, 6, 8), datetime.date(2015, 6, 10), datetime.date(2015, 6, 12), datetime.date(2015, 6, 16), datetime.date(2015, 6, 18), datetime.date(2015, 6, 22), datetime.date(2015, 6, 24), datetime.date(2015, 6, 26), datetime.date(2015, 6, 30)]

daily trading volume graph over a month

data = get_pricing('SPY', start_date='2015-6-1', end_date='2017-7-1', fields='volume')

daily trading volume over 2 years

data = get_pricing('SPY', start_date='2015-6-1', end_date='2015-6-2', frequency='minute', fields='volume')
...
for i in range(0, tick_num):
x_label.append(data.index[int(x_tick[i])])

[ 1130598. 860265. 466341. 305119. 420599. 434213. 456097.
485796. 321990. 256441. 276184. 193212. 286944. 167065.
108485. 197818. 422527. 593151. 479281. 214867. 601252.
268215. 570393. 303907. 262587. 203801. 176336. 654809.
242458. 227793. 854860. 557071. 1116683. 310044. 356430.
...
100768. 366257. 81084. 184404. 86525. 167199. 196828.
111246. 203550. 299601. 147401. 122207. 235673. 216496.
278106. 352249. 1112440. 418138. 273468. 270791. 439510.
424709. 117943. 172694. 181214. 317756. 765112. 316917.
522166. 798810. 1502124.]
DatetimeIndex(['2015-06-01 13:31:00+00:00', '2015-06-01 13:32:00+00:00',
'2015-06-01 13:33:00+00:00', '2015-06-01 13:34:00+00:00',
'2015-06-01 13:35:00+00:00', '2015-06-01 13:36:00+00:00',
'2015-06-01 13:37:00+00:00', '2015-06-01 13:38:00+00:00',
'2015-06-01 13:39:00+00:00', '2015-06-01 13:40:00+00:00',
...
'2015-06-02 19:51:00+00:00', '2015-06-02 19:52:00+00:00',
'2015-06-02 19:53:00+00:00', '2015-06-02 19:54:00+00:00',
'2015-06-02 19:55:00+00:00', '2015-06-02 19:56:00+00:00',
'2015-06-02 19:57:00+00:00', '2015-06-02 19:58:00+00:00',
'2015-06-02 19:59:00+00:00', '2015-06-02 20:00:00+00:00'],
dtype='datetime64[ns, UTC]', length=780, freq=None)
[ 0. 65. 130. 195. 260. 325. 390. 455. 520. 585. 650. 715.]
[Timestamp('2015-06-01 13:31:00+0000', tz='UTC'), Timestamp('2015-06-01 14:36:00+0000', tz='UTC'), Timestamp('2015-06-01 15:41:00+0000', tz='UTC'), Timestamp('2015-06-01 16:46:00+0000', tz='UTC'), Timestamp('2015-06-01 17:51:00+0000', tz='UTC'), Timestamp('2015-06-01 18:56:00+0000', tz='UTC'), Timestamp('2015-06-02 13:31:00+0000', tz='UTC'), Timestamp('2015-06-02 14:36:00+0000', tz='UTC'), Timestamp('2015-06-02 15:41:00+0000', tz='UTC'), Timestamp('2015-06-02 16:46:00+0000', tz='UTC'), Timestamp('2015-06-02 17:51:00+0000', tz='UTC'), Timestamp('2015-06-02 18:56:00+0000', tz='UTC')]

minute trading volume over 2 days

data = get_pricing('SPY', start_date='2015-6-1', end_date='2015-6-2', frequency='minute', fields='volume')
data = data.resample('60T').sum()

[ 11360751. 18557297. 7024886. 7016271. 5862379. 7315033.
17311150. 1869567. nan nan nan nan
nan nan nan nan nan nan
nan nan nan nan nan nan
16010783. 14301804. 5977053. 10200373. 4987498. 8257317.
14418781. 1502124.]
DatetimeIndex(['2015-06-01 13:00:00+00:00', '2015-06-01 14:00:00+00:00',
'2015-06-01 15:00:00+00:00', '2015-06-01 16:00:00+00:00',
'2015-06-01 17:00:00+00:00', '2015-06-01 18:00:00+00:00',
'2015-06-01 19:00:00+00:00', '2015-06-01 20:00:00+00:00',
'2015-06-01 21:00:00+00:00', '2015-06-01 22:00:00+00:00',
'2015-06-01 23:00:00+00:00', '2015-06-02 00:00:00+00:00',
'2015-06-02 01:00:00+00:00', '2015-06-02 02:00:00+00:00',
'2015-06-02 03:00:00+00:00', '2015-06-02 04:00:00+00:00',
'2015-06-02 05:00:00+00:00', '2015-06-02 06:00:00+00:00',
'2015-06-02 07:00:00+00:00', '2015-06-02 08:00:00+00:00',
'2015-06-02 09:00:00+00:00', '2015-06-02 10:00:00+00:00',
'2015-06-02 11:00:00+00:00', '2015-06-02 12:00:00+00:00',
'2015-06-02 13:00:00+00:00', '2015-06-02 14:00:00+00:00',
'2015-06-02 15:00:00+00:00', '2015-06-02 16:00:00+00:00',
'2015-06-02 17:00:00+00:00', '2015-06-02 18:00:00+00:00',
'2015-06-02 19:00:00+00:00', '2015-06-02 20:00:00+00:00'],
dtype='datetime64[ns, UTC]', freq='60T')
[ 0. 2.66666667 5.33333333 8. 10.66666667
13.33333333 16. 18.66666667 21.33333333 24. 26.66666667
29.33333333]
[Timestamp('2015-06-01 13:00:00+0000', tz='UTC', offset='60T'), Timestamp('2015-06-01 15:00:00+0000', tz='UTC', offset='60T'), Timestamp('2015-06-01 18:00:00+0000', tz='UTC', offset='60T'), Timestamp('2015-06-01 21:00:00+0000', tz='UTC', offset='60T'), Timestamp('2015-06-01 23:00:00+0000', tz='UTC', offset='60T'), Timestamp('2015-06-02 02:00:00+0000', tz='UTC', offset='60T'), Timestamp('2015-06-02 05:00:00+0000', tz='UTC', offset='60T'), Timestamp('2015-06-02 07:00:00+0000', tz='UTC', offset='60T'), Timestamp('2015-06-02 10:00:00+0000', tz='UTC', offset='60T'), Timestamp('2015-06-02 13:00:00+0000', tz='UTC', offset='60T'), Timestamp('2015-06-02 15:00:00+0000', tz='UTC', offset='60T'), Timestamp('2015-06-02 18:00:00+0000', tz='UTC', offset='60T')]

hourly trading volume in 2 days

reference:
https://www.quantopian.com/lectures/introduction-to-volume-slippage-and-liquidity
https://pythonspot.com/matplotlib-bar-chart/
https://matplotlib.org/3.1.1/api/_as_gen/matplotlib.pyplot.xticks.html
https://pandas.pydata.org/pandas-docs/stable/reference/api/pandas.DataFrame.resample.html

Saturday, 7 December 2019

matplotlib area fill

import numpy as np
from scipy import stats
import matplotlib.pyplot as plt

x = np.linspace(-5,5,100)
#normal distribution, mean =0, deviation = 1
y = stats.norm.pdf(x,0,1)
plt.plot(x,y)

plt.fill_between(x, y, color='yellow')

fill between normal distribution and x axis

y2 = x**2
plt.plot(x,y2)

plt.fill_between(x, y, y2, color='yellow')

plt.ylim(0, 0.4)

fill between normal distribution and parabola

plt.fill_between(x, y, y2, where=y>y2, color='yellow')

filter out outside region

reference:
https://www.quantopian.com/lectures/confidence-intervals
https://www.science-emergence.com/Articles/How-to-fill-an-area-in-matplotlib-/
https://matplotlib.org/3.1.1/gallery/lines_bars_and_markers/fill_between_demo.html

Friday, 6 December 2019

哪些国外经历回国可能会大幅降薪, 哪些海外工作经历回国高薪？

第一，中国比国外还领先的行业。

比如说，有同学在日本留学，毕业之后在日本的电商网站工作，工作一段时间之后想回国，觉得自己有在海外的工作经验，所以找工作开价很高，但是一直没能找到如愿的工作。实际上，中国的互联网，电商行业已经处于全球领先的地位，国内并不缺乏相关的人才，所以在海外的电商经验，并不能成为回国找相关工作的助推器。还有人在新加坡做APP项目运营，新加坡的APP水准是远落后中国，这样的经历回国后，如果直接按照汇率定价很高，都比较难找到理想工作。

第二，还有一些中外行业差别巨大的工作，国外经历用处不大。

比如税务、银行。举个例子，有同学在国外读会计专业，在国外的会计事务所有税务经验，回国想做金融，但是工资要求和国外持平或者更高，这一点实际很难做到。每个国家的税法都是不同的。

关于会计岗位，中国不缺初级、中级财务人才。如果国外比较底层的经验，不太可能空降成为高薪的财务经理等职位。特别是在澳大利亚学习会计的同学，有着相当大的比例，尤其需要注意这一点，并不是有了国外的普通会计经验，就可以回国拿高薪，除非你已经做到CFO的级别，那很缺。更何况国内的会计已经是趋于饱和，想找一个懂中国会计法和税法的熟手并不是难事。

银行属于极度完善和流水作业的行业，越发达的银行业，流水线程度越高。如果只是在国外银行某个环节搭把手干活。回来想拿到很高的薪水，非常难。甚至找个特别满意的工作都比较难。

第三，国外有而中国无的职位，回国很难找工作。

有同学在英国留学，学习会计金融，在英国的银行有外汇清算的工作经验。但是回国很久都找不到工作。原因在于，英国有很开放的外汇交易，所以需要外汇清算这个岗位，但是中国目前的汇市并不开放，能够提供外汇清算的岗位少之又少，所以找不到相关工作是必然。欧洲作为工业革命的起源地，在很多行业都遵循着行业流水线化的原则，经过多年的发展，很多成熟行业的工作已经流水化作业，每个人负责的部分是有限的一小块。但是在中国，部分行业的发展还没有成熟到流水线化，有些在欧洲有的岗位，在中国未必能找得到。

第四，技术类的基础岗。

因为华人在国外，要找个高起点的工作比较难。所以绝大部分工作经验在5年以内的国外经验都还处于边缘岗位，工作经验对比国内同龄人逊色一大截。

有的同学在国外做了一年数据分析，或者国外做了一年四大审计，希望回国找个2-3万的工作。这个是比较难的。

因为数据分析本身是一个入门性质的岗位，一方面中国还没发达到基础入门岗给2-3万，另一方面，中国数据方面人才真心不缺！只会比国外的好！

简历君，见过有人在美国咨询公司做了5年数据分析回国。一开始定位年薪60-70万。最后找了七八个月工作都没着落。朋友给推荐了去BAT，结果面试官一问，“啊，你在国外做的这个很基础啊，我们应届生都可以做。我们做得这个是人工智能水准了，你胜任不了。”最终市场给了1.5万月薪。

而四大审计，本身审计就不是什么高薪高难度工作。中国审计行业发展几十年了，多牛的人都一大把。哪个国家四大回来的人都一堆。所以国外四大回来，二十年前会有个比较不错的薪水。放在21世纪，可能要小小失望一把。

哪些海外工作经历回国高薪呢？

通用技术型工作是可以在国外镀金的，比如医生，金融量化交易、国外做过国际大企业总部品牌营销，美国娱乐业，硅谷高科技公司重要项目负责人。此类工作属于通用型的工作，并不会因为国家不同而有太大差别，而且国外技术比中国更成熟。当然，一样，要国外的高级人才回来。这一类的工作，不仅回国会比较吃香，就算是想留在国外也是比较好找工作的，薪水也普遍不错。

Thursday, 5 December 2019

quantopian curve fit gaussian + polynomial

import numpy as np
import matplotlib.pyplot as plt
from scipy.optimize import curve_fit
import pandas as pd

def expfit(x, a, b, c, d, e, f, g):
return a*np.exp(-(b*x*x+c*x+d))+e*x*x+f*x+g

x = np.linspace(1,1+len(price),len(price))

params, params_covariance = curve_fit(expfit, x, price.values )

print params

y = expfit(x, *params)

f = pd.Series(y, index=price.index, name='exp fit')

pd.concat([price, f], axis=1).plot()

[ 1.87976840e+03 -2.61667985e-06 6.40016840e-03 3.83225576e+00
-1.06443204e-05 7.86290764e-02 -1.36885462e+01]

[ 1.15483490e+01 1.10677384e-04 -1.16953358e-02 -1.34140290e+00

-7.36267261e-06 3.04019894e-04 5.55989990e+01]

[ 1.58855939e+02 9.57979344e-05 -1.48746514e-02 1.72812129e+00

-4.63387042e-05 1.52436240e-01 3.93288443e+01]

def expfit(x, a, b, c, d, e, f, g, h):

return a*np.exp(-(b*x*x+c*x+d))+e*x*x+f*x+g+h*x*x*x

[ 7.19650057e+02 2.88317903e-06 2.24186888e-03 7.27803145e-01

-5.10066383e-04 8.24549494e-01 -2.58031805e+02 9.90353284e-08]

gaussian + 3rd order polynomial

[ 9.86790120e-03 9.34621548e-05 -9.37662256e-03 -8.68029637e+00

-5.06336033e-05 5.08207847e-02 4.03453234e+01 1.05790726e-08]

reference:
http://chuanshuoge2.blogspot.com/2019/12/quantopian-curve-fit-gaussian-linear.html

quantopian curve fit gaussian + linear

start = '2008-01-01'
end = '2018-01-01'
price = get_pricing('AAPL', fields='price', start_date=start, end_date=end)
price.plot()

import numpy as np
import matplotlib.pyplot as plt
from scipy.optimize import curve_fit
import pandas as pd

def expfit(x, a, b, c, d, e, f):
return a*np.exp(-(b*x*x+c*x+d))+e*x+f

x = np.linspace(1,1+len(price),len(price))

params, params_covariance = curve_fit(expfit, x, price.values )

print params

y = expfit(x, *params)

f = pd.Series(y, index=price.index, name='exp fit')

pd.concat([price, f], axis=1).plot()

[ 7.84216333e+01 5.57230815e-05 -7.13434301e-03 1.45807761e+00
5.84734404e-02 -4.80640750e+00]

[ 8.44865621e+00 1.29652892e-04 -1.37545875e-02 -1.42288153e+00

-1.96774039e-02 6.58394726e+01]

reference:
https://chuanshuoge2.blogspot.com/2019/12/quantopian-curve-fit-gaussian.html

quantopian curve fit gaussian

start = '2008-01-01'
end = '2018-01-01'
price = get_pricing('XOM', fields='price', start_date=start, end_date=end)
price.plot()

import numpy as np
import matplotlib.pyplot as plt
from scipy.optimize import curve_fit
import pandas as pd

#Gaussian formula

f(x)=ae^{-{\frac {(x-b)^{2}}{2c^{2}}}}

g(x)={\frac {1}{\sigma {\sqrt {2\pi }}}}e^{-{\frac {1}{2}}\left((x-\mu )/\sigma \right)^{2}}.

def expfit(x, a, b, c, d, e, f):
return a*np.exp(-(b*x*x+c*x+d))+e

x = np.linspace(1,1+len(price),len(price))

params, params_covariance = curve_fit(expfit, x, price.values )

print params

y = expfit(x, *params)

f = pd.Series(y, index=price.index, name='exp fit')

pd.concat([price, f], axis=1).plot()

[ -4.26776636e-01 3.41090632e-06 -3.42580712e-03 -3.33515628e+00
8.08866791e+01 1.00000000e+00]

reference:
https://en.wikipedia.org/wiki/Gaussian_function
http://chuanshuoge2.blogspot.com/2019/12/python-curve-fitting.html
http://chuanshuoge2.blogspot.com/2019/12/quantopian-curve-fitting-log.html