chrisnatali / finx Goto Github PK

Market Data Gathering, Analysis and Trading application.

Dependencies: R, Ruby

Load:
- Scripts for gathering market data from the web

r:
- Scripts for analysis and trading simulation

Setup:
- Set the FINX_HOME env variable to the home directory
  of your FinX environment 
  (i.e. export FINX_HOME=/home/joe_schmoe/FinX)

- cd to the load directory

- Retrieve stock data from the web

-- Run ./scripts/stock_set_update.sh
   This loads the set of stocks (without their data)

-- Run ./scripts/stock_detail_update.sh tick-cap
   This gets the detail data for each stock of interest
   (currently only the dividend rate)

-- Run ./scripts/stock_history_update.sh tick-cap
   This gets daily trade data going back to 2000 for all
   stocks in the set retrieved earlier.

- Load R environment with stock data

-- Run R (from the load directory)

-- execute load-tdt.R via:  load('../source/load-tdt.R')
   This loads the ticker history into one large dataframe,
   and adds some calculated columns to it.

-- Use buy-sell.R as template for selecting sets of tick/day
   combinations to buy or sell and store them in dataframes.

-- Use functions in backtest.R to simulate trading those
   tick/day combinations.  Pass the dataframes from prior step
   into bt.sim to simulate trading those positions over the
   history of market data.  Check bt.pool and bt.ledger for
   results.  

-- Plot, iterate and analyze as necessary

TODO:
- Simulate shorts & test
- Elaborate & Refine functionality (i.e. Run nightly & emit buys/sells)
- Run nightly on home Linux machine (add RAM)
- Formalize abstraction into DDL
- Create market index to relate other data to.  
  - Retrieve outstanding shares data
  - Pull in ticker level data to R data frame
  - Calculate total value of all shares (sum(tdt$cls * tick$shares))
    per day, then chg variable for this index
- Add varying trend durations and analyze which work best 
  over different periods (and try to categorize those periods 
  based on market index)
- Add clustering of ticks based on trend information.  Include
  aggregate cluster information into analysis (i.e. if this 
  stock belongs to a LARGE cluster, then favor it over one in
  a small cluster).  
- Rather than a random sample of those stocks selected via a signal,
  order them by how closely they match the signal.