au_finance

Inhouse tools and database for computing and visualizing some metrics in the Australian financial market.

interest_rate_model

In this section, the linear model for exchange rates between two arbirary countries is examined using the historical data for US and Australia.

interest_exchange_rate.db is a SQLite database that contains the exchange between AUD and other currencies and interest rates of Australia and US up to Jun 2016.

interestRateAnalyticalModel.py uses the monthly data Year 1997 and the analytical model below,

$$ \mathrm{ln} \frac{S_n}{S_{n+1}} = \alpha (i_n^{AU} - i_n^{US}) + \beta $$

where $i$ and $S$ is the interest rate and the exchange rate, respectively. $n$ denotes a period and in this particular case, is month.

From the figure above, we can see the strong correlation between the interest rate and the exchange rate, despite the fluctuations in the function of the exchange rate function $\mathrm{ln} \frac{S_n}{S_{n+1}}$. One apparent discrepancy is observed in the year 2008 when a global economic crisis occurred.

stock

yahoo_finance.py contains useful functions to grab data from and return the result as pandas.DataFrame.

pplot can visualise the OHLC and volume data retrieved from Yahoo Finance and analyse with the moving average.

ASXListedCompanies.pckl is a pandas.DataFrame that contains the names, symbols and the sectors of the listed companies.

ASXListedCompanies.xlsx is the spreadsheet for the same data with a pivot table for different sectors.

def get_histdata(symbol,startDate="1990-01-01",endDate=None)

Grab historical data for a single stock using its code or symbol

def get_selected_hist(stocks)

Grab historical data for some stocks using a list of codes or symbols

market_structure.py contains functions to visualize market structures of a collection of stocks. This module requires sklearn (Python machine learning package) to categorise market data. This idea here is to use the daily variation of individual stocks and find out which stocks are usually varying together. By calling covariance.GraphLassoCV, we obtain a graphical represenation of the correlation patterns. Then, with cluster.affinity_propagation the program automatically decides the number of clusters and print out the results. To visualize the data, we choose to use a 2D representation because it is more human-friendly and with manifold the dimension of the data is reduced to 2D.

def discover_clusters(var)

input the covariance and return labels of clusters and partial correlations that can be used in plotting

def get_low_dim_embedding(X, n_neighbors=4)

get a low dimension representation using X as the N by N matrix

def plot_market_structure(names, labels, embedding, partial_correlations)

the plotting function

## dependencies *numpy, pandas and scikit-learn*