Many people have attempted to become the next great trader. They try assets like cryptocurrencies, or futures, or options and soon find out that it isn’t as easy as they originally thought. They run into many of the same problems. … Success comes with diligent work, support, and access to mentors, technology, and data.

Successful Starts in Algorithmic Trading

Getting Started in Algorithmic Trading Includes Overcoming Common Problems

Feb 24, 2018

by Tayloe Draughon, Product Manager, and FINTECH veteran

Many people have attempted to become the next great trader. They try assets like cryptocurrencies, or futures, or options and soon find out that it isn’t as easy as they originally thought. They run into many of the same problems.

Problem: Not enough cash.

I recently lost some personal cash on short-term volatility. I knew it was a gamble. I should have sold my short volatility ETF prior to the presidential state of the union speech. I lost. Oh well; it happens. It was a small position. Most people starting out can’t afford even a small loss. Any speculative trading has the possibility of a loss.
People need capital to trade. If you don’t have it, then you need access to someone else’s capital, which usually comes with a professional trading position. This presents another problem. If you aren’t already a professional trader then leaving your job to become a full-time trader that is paid entirely out of profits is fear-worthy.

Problem: Not enough technology.

To build a trading system one needs computers that serve upmarket data, order processing to a broker or directly to the exchanges, risk systems, monitoring systems, data stores, and trade reconciliation systems. Once all of this is in place then you have to build an algorithmic trading platform that interfaces with all of these trading components. Also, you can use a third-party trading platform but this requires cash too.

Problem: Not enough data.

There are multiple types of data. Modern trading signals, including alpha signals, are derived from market data, fundamental data, news, social sentiment, and other “alternative data” (Alt-Data) sets.(1) The majority of trading algorithms require historical and streaming data. Historical data for the development phase and live data for trade execution. Most quantitative developers have an idea for their strategy. To figure out if it will actually work they will need access to the data to analyze, backtest, and forward test prior to beginning to trade.

Problem: Expensive data.

Market data is expensive. Access to the data is governed by the data owner. You can use it to trade with a retail account. However, access to the data for your research and analysis is a much bigger problem. The exchanges require legal agreements and charge for data.
Fundamental data that provides earnings, calendars, estimates and corporate information is provided by other data vendors. They too require legal agreements and for you to contribute to their revenue stream.
Alt-Data is also expensive and requires another paid licensing agreement. A recent conference pointed out that there are roughly 1,600 alternative data set providers.
Lastly, you have to manage all this data. Storage, organization, and formatting all become an issue. One data source may use exchange symbols and another may use an internal identifier. You have to spend time and money storing and formatting the data.

Problem: Not enough protection of intellectual property

Your trading strategy is your intellectual property. Proprietary trading companies run custom algorithms every day. Quants entering into algorithmic trading need to be aware that their intellectual property may not actually be theirs. For example, quants at trading firms get a salary and in return, the employer has justifiable rights to the work their employees produce.
Those that use free services to develop algorithms need to read and understand any user license agreement. (Don’t just click through!)

Problem: Limited understanding of trade expression

The markets don’t always work the way that an algo developer expects. Order types, time of day, order execution strategies, exchange rules, broker supplied algorithms, and a myriad of other order transaction items may cause problems for the algo creator. These items hinder the development of a quantitative algorithm because they distract the data scientist from studying trading strategies and may result in curve fitting.

Problem: Not enough mentoring or coaching

If good trading comes from experience then experience must come from bad trading. There is a reason we backtest and paper trade. Over time one gets better as one perfects one’s trading strategies. Traders develop operational “golden rules” that work and signals that improve their trading. They are constantly looking to improve.
In algorithmic trading, the role of a mentor or coach helps develop these disciplines and helps you find new opportunities. If your algorithmic strategy development exists in a vacuum then it is likely to perform poorly. A coach or mentor helps point out new data sets, new ideas, new risk tools that you can leverage to create a fresh, proprietary strategy that is your unique property.

Overcoming the problems

For a retail trader or someone entering the market, these problems can be daunting. Rest assured, there are solutions. For the (highly) capitalized trader, these problems can be used as a roadmap to build a fully functioning algo development and trading environment. Solutions for those with capital depend upon the asset class being traded. When evaluating your choice of systems and approaches consider looking for technology that will allow you to be broker-neutral.
Many (less capitalized) are coming to CloudQuant (where I work). The free tools allow you to test and build algo trading signals from market data, alt-data, and fundamental data. Signals can be joined together into a trading strategy. A strategy can be backtested to generate highly insightful results. The goal is to convert crowd-sourced algos into funded and licensed trading strategies that can be traded with our capital.
Even with barriers lowered, algo development is still hard work
A trading algorithm takes time to develop. Most people get stuck quickly and even give up. When you get stuck, or your algo doesn’t show progress, you need to stop and re-evaluate. An approach that works well within most algo shops is to break the strategy down into small components.
Consider these steps:

  1. Identify a data feature.
  2. Repeat #1 as many times as possible
  3. Look for ways to combine your features into something that you can use to predict a price movement.
  4. Then test your algo

When you are stuck, ask your coach or your community for ideas. People are willing to help you out.
Success comes with diligent work, support, and access to mentors, technology, and data.

1. Alternative Datasets are data sets that contain information not normally used in trading. These may be social or news sentiment analysis, satellite data, traffic data, or other information that is available. Natural Language Processing has created an explosion of new data sources for the imaginative quant.