Peter Berg

Created a framework for converting historical price data in virtually any format to another format of our choosing.

The utility was also capable of simultaneously filtering the data being converted (using an optionally specified filter) and synthesizing it into OHLC bars (using an optionally specified synthesizer).

Compatible forms of input data:

-Binary files
-Delimited or padded text files
-Tick data
-OHLC data
-Daily Data
-Intraday Data
-Stock/Futures… Show more

Created a framework for converting historical price data in virtually any format to another format of our choosing.

The utility was also capable of simultaneously filtering the data being converted (using an optionally specified filter) and synthesizing it into OHLC bars (using an optionally specified synthesizer).

Compatible forms of input data:

-Binary files
-Delimited or padded text files
-Tick data
-OHLC data
-Daily Data
-Intraday Data
-Stock/Futures Data
-Data for Instruments priced in foreign currencies
-Data for Instruments with stock splits

My role in this project focused on developing the needed abstractions to achieve this flexibility, as well as integrating our filtering and ohlc bar synthesis logic.
Show less

Created a modified version of our existing optimization abstraction designed to be able to run over extreme market conditions (i.e. highly volatile periods) without being disproportionately affected by the results during those periods.

Created a library enabling users to create order prioritization systems. These systems were for controlling situations in which algorithms were generating more orders than could be placed due to capacity constraints.

Re-architected an antiquated portion of the code base to make algorithm results discrete from the objects representing the algorithms themselves and leveraged this abstraction to create a system for writing the results of our trading algorithms to file and then reading them back into memory.

I designed and implemented our MultiVar object/protocol which makes a single run capable of reporting results for any number of parameter permutations.

MultiVars can be used to vary anything from a simple numeric threshold to complex conditions within algorithms.

Eliminated roughly 10,000 lines of code. Changed the API substantially to make it much more intuitive. Increased the speed of our backtester by about 2-3 times.

Developed a pricing mechanism for spreads and baskets allowing them to be treated as interchangeable with ordinary equities and derivatives. This enabled us to run algorithms on either single instruments or spreads/baskets without having to alter their code.

Pricing spreads was a particularly difficult issue because there is no guaranteed relation between the PnL of a spread and its cost (viz. it's possible for an x-dollar drop in the price of a spread to precipitate a profit of… Show more

Developed a pricing mechanism for spreads and baskets allowing them to be treated as interchangeable with ordinary equities and derivatives. This enabled us to run algorithms on either single instruments or spreads/baskets without having to alter their code.

Pricing spreads was a particularly difficult issue because there is no guaranteed relation between the PnL of a spread and its cost (viz. it's possible for an x-dollar drop in the price of a spread to precipitate a profit of anything between -x and x dollars). Show less

Developed a library for making quote filters and used the library to make two filters, one for intraday and one for daily quotes. Filters made using the library automatically work with both live and simulated data feeds. They handle latency issues when filtering data coming from live feeds.

Using JFreeChart, put together a simple framework for plotting relationships between price moves and user specified variables/statistics that was integrated with our algorithmic backtester. Was able to display Histograms, Scatter Plots, and Time Series.

Also included stackable "Transformers" for manipulating data before it was plotted.

This framework could be used to do something like look at the relationship between the normalized overnight move of all US stocks with average… Show more

Using JFreeChart, put together a simple framework for plotting relationships between price moves and user specified variables/statistics that was integrated with our algorithmic backtester. Was able to display Histograms, Scatter Plots, and Time Series.

Also included stackable "Transformers" for manipulating data before it was plotted.

This framework could be used to do something like look at the relationship between the normalized overnight move of all US stocks with average daily dollar volume of 5 million or more (~2000 symbols) and the open to close move on the following day over a period of two years. Coding this up using the framework (in conjunction with other pieces of the program like the SymStat library (see above)) would be trivial -- a two minute process. Show less

Built out a testing framework designed to assure that various components of the software produced identical results between builds. The framework operated on

Trading Algorithms
SymStats
ResultStats
Optimization Algorithms
Risk-Management Systems
Data Filter Modules
OHLC-bar Synthesizers

Created a system for managing risk across multiple algorithms being traded simultaneously. The system can be used to

-Ensure that net and gross position limits for single instruments and across asset classes are enforced
-Cancel the trades of under-performing algorithms
-View the combined performance of algorithms

Created technology to automatically seek out the optimal value of a parameter for a given algorithm. Structured the library in such a way that users could create their own optimization algorithms, in addition to using Georgica's.

SymStats are statistics for individual instruments calculated over nothing but historical price data. I designed the SymStat library to:

a) Make it so that commonly used metrics and indicators could be discretely defined in their own objects and easily referenced in algorithms
b) Save resources during large runs in which the same SymStat is being used in multiple algorithms (SymStats are internable).
c) Enable us to package valuable SymStats (including both proprietary formulas… Show more

SymStats are statistics for individual instruments calculated over nothing but historical price data. I designed the SymStat library to:

a) Make it so that commonly used metrics and indicators could be discretely defined in their own objects and easily referenced in algorithms
b) Save resources during large runs in which the same SymStat is being used in multiple algorithms (SymStats are internable).
c) Enable us to package valuable SymStats (including both proprietary formulas and widely used metrics) with our distribution of the software Show less

The ResultStat library enables users to customize their own results reporting by creating their own ResultStats and ResultStatSets. A ResultStat is calculated over a set of trades.

Eliminated 3,000 lines of code. Created several new behind the scenes abstractions enabling us to use the same classes to both run our algorithms and filter our data. Did this all without altering the API.