ffn - a financial function library for Python

I am running the calc_stats() function on my data frame whihc seems to be formatted correctly and has no errors ( i tried it on the 297 lines of data then 50 then 10, errors on all the different frames), i get the following error: C:\ProgramData\Anaconda3\lib\site-packages\ffn\core.py:2054: RuntimeWarning: invalid value encountered in minimum negative_returns = np.minimum(returns, 0.)

I am not very proficient in Python but i have done my best to see whats wrong with the data and it all looks clean. I am not using Yahoo data as the data retrieval that ffn is using also fails me, so I created my own data from elsewhere. Pls help as this would be very valuable to my project.

Thanks

When I import fan I have this error. I've tried to update fun.

`/opt/anaconda3/lib/python3.8/site-packages/ffn/core.py:27: MatplotlibDeprecationWarning:

The 'warn' parameter of use() is deprecated since Matplotlib 3.1 and will be removed in 3.3. If any parameter follows 'warn', they should be pass as keyword, not positionally.`

As per issue https://github.com/pmorissette/ffn/issues/10.

For the Calmar Ratio I used the method discussed here: http://investexcel.net/calmar-ratio/ For the Sortino Ratio I used this method: http://investexcel.net/calculate-the-sortino-ratio-with-excel/

I think I've implemented them correctly, but it might need a skilled eye to sense check it :)

Pandas 1.0.0 has removed the DataFrame.ix indexer. This indexer was being used in several core functions including rebase() and calc_cagr(). In all, there are 11 instances of this indexer being used in the core module. Pandas also seems to have become aggressive in their policy of requiring users to update. It would be good to update the indexer to the suggested .iloc or .loc, as appropriate.

Maybe we should just use empyricaland pyfolioinstead of ffn. @pmorissette @FinQuest What do you guys think?

They seem similar and they have a strong community already. empyrical can replace the generic functions in ffn and pyfolio can replace PerformanceStats. We could still use ffn for calculations that they don't include like calc_mean_var_weights or calc_erc_weights but we wouldn't have to worry about maintaining the portfolio statistics or plots. Over time we could see about contributing any functionality they don't include into their libraries.

I still prefer bt over zipline because it is easier to understand and verify the functionality is correct. Trusting the backtesting tool is essential if you are going to use the results to put money to work.

Hi,firstly i want to say thank you! this is an awesome package for financial researchers.

This is my code -

from pandas.io.data import DataReader import ffn moat=DataReader('MOAT','yahoo').reset_index().loc[:,['Date','Adj Close']].set_index('Date') moat=moat[moat.index<='31-aug-2014'] moat.tail() ffn.core.to_monthly(moat).tail()

my question is, why do we not see the August data after converting to monthly?

Hi all,

Two days ago, I tried to use ffn to get data from yahoo finance (by default), but the program left this error message to me.

My statement was:

aapl = ffn.get("aapl", start = "2020-05-06")

The program showed me:

RemoteDataError: Unable to read URL: https://finance.yahoo.com/quote/aapl/history?period1=1588737600&period2=1625371199&interval=1d&frequency=1d&filter=history

I have tried different ticks and doublechecked the usage manual to see if any API is changed, but in vain. So would you please help me solve this issue? Thank you.

BR, Arthur

Fixed bug in YTD and MTD calculation. YTD and MTD are not calculated for short return series, and are not necessarily equal to total_return.

MTD is only assigned something other than total_return, if more than one monthly return exists - which is unnecessary.

Same issue for YTD - but with a stricter condition of more data before it is calculated.

MTD and YTD returns are calculated after validating that a single daily return is available. If in the same month, the dp series can be used to find MTD, else the second to last value in the mp series is used. Same method for YTD calculation.

Hi,

Firstly, LOVE this package - it has been a god send to me so many thanks for your hard work and choosing to share it with us all.

I have noticed what I believe to be a "bug" with the ".monthly_returns" when used on certain input data and with a PerformanceStats object.

When using daily periodicity data (or intra-day) as a base input to the PerformanceStats object, if there is a whole calendar month period without a corresponding entry in the series (or more than one month in a row), the resulting data returned seems to be incorrect/incomplete.

I have provided two csv input files and a Jupyter notebook that replicates the 2 issues I have found, with further explanation within.

I may be missing something my end which is causing this, but I do believe it to be a legitimate "bug". ffn_bug_examples.zip

pandas-datareader 0.7.0 (released today) fixes the latest Yahoo data breakage, but Yahoo is now using new field values which breaks ffn.get(). For instance, 'Adj Close' changed to 'adjclose'. I replaced the two references to 'Adj Close' with 'adjclose' in data.py and now ffn.get() returns data, but the dataframe is only number indexed rather than date indexed. It's my first day with ffn so I'm not sure what typical behavior is, but I was just comparing my results to the examples and saw this discrepancy.

Hi

Today, when I was using ffn, I found that an error would be reported. After checking, it should be that the information obtained by yahoo finance has been revised. Could you please check it?

Here is the discussion I found https://stackoverflow.com/questions/74832296/typeerror-string-indices-must-be-integers-when-getting-data-of-a-stock-from-y

thank you very much

Hi,

Im using bt for backtesting (great tool btw) and im exploring crypto data which trades 365 days a year. I made ffn.core. TRADING_DAYS_PER_YEAR equal to 365 and when I ran my backtests which have 5 years of daily data, I didnt see any differences in the performance stats. I assumed the annualized returns at least would change. So I have two questions...

1. what performance metrics should change when I change TRADING_DAYS_PER_YEAR from 252 to 365?
2. The docs say use 360 for crypto. Why would it be 360 and not 365?

Thanks!

Evaluate what functionality empyrical has that ffn is missing and consider implementing as empyrical is not really maintained.

https://github.com/quantopian/empyrical https://github.com/quantopian/empyrical/issues/125

Line 2302:

res = np.divide(er.mean(), std)

Unless I'm reading this incorrectly, this is using the mean instead of geometric mean which is incorrect (should be .gmean(), right).

            self.daily_mean = r.mean() * 252
            self.daily_vol = np.std(r, ddof=1) * np.sqrt(252)

            # if type(self.rf) is float:
            if isinstance(self.rf, float):
                self.daily_sharpe = r.calc_sharpe(rf=self.rf, nperiods=252)
                self.daily_sortino = calc_sortino_ratio(r, rf=self.rf, nperiods=252)
            # rf is a price series
            else:
                _rf_daily_price_returns = self.rf.to_returns()
                self.daily_sharpe = r.calc_sharpe(
                    rf=_rf_daily_price_returns, nperiods=252
                )
                self.daily_sortino = calc_sortino_ratio(
                    r, rf=_rf_daily_price_returns, nperiods=252
                )

The yearly return data is hard coded to 252 days which accurately represents trading on the NYSE or other standard markets. However, other markets trade 365 (366) days per year, or somewhere in between (e.g. 6 days a week for futures). The returns should either dynamically adjust based on the time series presented or offer an ability to set the number of trading days.

I used monthly data and checked the performance. I always come to the result that with "six_month" the last 7 lines are calculated. otherwise, I do not come to the same result. or am I wrong, or maybe monthly data are not suitable.