search

pat42w / EF_Portfolio_Optimization

This project aims to test Portfolio Optimization methods on stock data in python.
MIT License
2 stars 0 forks source link

Convert all gains from index currency to Euro #4

Closed pat42w closed 3 years ago

pat42w commented 3 years ago

Currently all portfolios are in the index's currency rather than euro.

Perhaps a database of daily Euro to other currency prices is the way to go here

eohara-ie commented 3 years ago

When would we want to do the conversion back to EUR? My guess would be the date that our Expected Returns prediction is for. Not something we could do for our future predictions.

Perhaps a similar algo is needed to predict FX rates? lol

pat42w commented 3 years ago

First step would be to have a historic price per euro of any currency we are using covering our date range , for testing. then I'd probably use the FB prophet model for time series to predict future exchange rates we can test it out and see ?

pat42w commented 3 years ago

The more i think on this would we want to be using closing price * exchange rate on a daily level here when running the Efficient frontiers this way we capture the aditional risk& variance due to operating in a different currency all in one ?

eohara-ie commented 3 years ago

Would be interesting to see how that affects results alright

eohara-ie commented 3 years ago

To do: Calculate covariance of volatility of EUR / USD pair

eohara-ie commented 3 years ago

Steps:

  1. Pull average daily USD price in EUR for time interval
  2. Calculate percentage change in USD price
  3. Calculate variance
  4. Volatility
  5. Covariance Nothing to check correlation of USD price with here? Could check correlation with some symbols later
  6. Expected Returns ~ Expected USD price (?)

Only real unknown is getting data

eohara-ie commented 3 years ago

https://forex-python.readthedocs.io/en/latest/usage.html

pat42w commented 3 years ago

We don't treat the currency as an asset in itself. Instead we are standardising our prices to euros. Prior to running our portfolio s

  1. Pull average daily USD price in EUR for time interval
  2. Create function currency_conversion()

Inputs:

Outputs:

eohara-ie commented 3 years ago

Two functions in DatabaseMainFnc.py:

gen_curr_csv:
param: currency pair (default [USD->EUR,JPY->EUR,GBP->EUR])

Generates dataframe for currency pair(s) between 2006 and yesterday, and saves to csv

and

load_curr_csv:
params: stocks_df
    input_curr
    output_curr

Loads currency csv
converts date col to index
Checks if input currency is in currency_df column names
Prompt user to run gen_curr_csv if not (or run automatically?)
If output currency is not EUR,
rates = conversion from output to EUR to input
If output is EUR,
rates  = column values where column = output

**return** rates \* stocks_df row-wise (index-wise?)
    vectorised anyway
eohara-ie commented 3 years ago

Committed changes to currency_dev branch. gen_curr_csv and load_curr_csv are now in. Multiplication using numpy looks to work as we want. Can check with the proper stocks dataframe.

A couple of changes from previous comment:

I hard-coded the 3 currency pairs in gen_curr_csv to simplify things.

I also fixed the output currency to EUR in load_curr_csv. If an input currency that isn't JPY, GBP or USD is passed in, it should return with a Currency not Supported message

Apart from that should be all g

I didn't properly look at your latest notebook on the main branch yet, can look at integrating after we merge these branches

Will make a merge request now

pat42w commented 3 years ago

Tried the currency conversion in Historic_portfolio_nb.ipynb & I get the following error:

#Test the currency conversion
dmf.load_curr_csv(df_Nasdaq['AACG'],'USD')

ValueError Traceback (most recent call last)

in 1 #Test the currency conversion 2 ----> 3 dmf.load_curr_csv(df_Nasdaq['AACG'],'USD') c:\Users\patri\OneDrive\Documents\GitHub\EF_Portfolio_Optimization\DatabaseMainFnc.py in load_curr_csv(stocks_df, input_curr) 304 305 # Multiply each row of stocks dataframe by its' corresponding exchange rate --> 306 result = pd.DataFrame(np.array(rates_df) * np.array(stocks_df),columns=stocks_df.columns,index=stocks_df.index) 307 308 return result ValueError: operands could not be broadcast together with shapes (5579,4) (3848,) Looks like a shape issue and its trying to use maybe the index needs to be the dates of prices database, and its trying to use all the currencies instead of just the input_curr?
eohara-ie commented 3 years ago

Ah yes, it should be rates_df[input_curr] on that line. I think it should work then

-------- Original Message -------- On 11 Apr 2021, 20:59, pat42w wrote:

Tried the currency conversion in Historic_portfolio_nb.ipynb & I get the following error:

Test the currency conversion

dmf.load_curr_csv(df_Nasdaq['AACG'],'USD')

ValueError Traceback (most recent call last) in 1 #Test the currency conversion 2 ----> 3 dmf.load_curr_csv(df_Nasdaq['AACG'],'USD') c:\Users\patri\OneDrive\Documents\GitHub\EF_Portfolio_Optimization\DatabaseMainFnc.py in load_curr_csv(stocks_df, input_curr) 304 305 # Multiply each row of stocks dataframe by its' corresponding exchange rate --> 306 result = pd.DataFrame(np.array(rates_df) * np.array(stocks_df),columns=stocks_df.columns,index=stocks_df.index) 307 308 return result

ValueError: operands could not be broadcast together with shapes (5579,4) (3848,)

Looks like a shape issue and its trying to use maybe the index needs to be the dates of prices database, and its trying to use all the currencies instead of just the input_curr?

— You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub, or unsubscribe.

pat42w commented 3 years ago

Ah okay, does that sort the length issue too?

eohara-ie commented 3 years ago

Ah I see now. Length issue will probably still be there. Do you know why the stocks_df is 3k long? Is it a new stock(i.e. joined the market after Jan 2006)? I hadn't considered that possibility.

Will need to do a join of some sort, keeping the intersection of the two indexes

-------- Original Message -------- On 11 Apr 2021, 21:25, pat42w wrote:

Ah okay, does that sort the length issue too?

— You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub, or unsubscribe.

pat42w commented 3 years ago

Well thered be currency rates for every day but stock data only on trading days . I'd say that's why it's shorter. But I think the datetime indexes can be used to take the dates needed ,

eohara-ie commented 3 years ago

Ah yes that too. Ok I'll add in that step before the multiply -------- Original Message -------- On 11 Apr 2021, 21:34, pat42w wrote:

Well there's be currency rates for every day but stock data only on trading days . I'd say that's why it's shorter. But I think the datetime indexes can be used to take the dates needed ,

— You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub, or unsubscribe.

eohara-ie commented 3 years ago

Ok if I make this patch directly onto main branch?