Best Automated Trading Platforms For Algo Trading

Algorithmic trading explained

Once in a while, the market will go through a period of turbulence that hit all quant funds in the same manner, leading to steep losses for all of them.

One such event was the Covid-19 pandemic in March 2020. Market turbulence dragged quant funds deep into the red (see below).

Source: Financial Times (paywall)

The best algorithmic trading platforms depending on what type of trader you are, professional traders will need a DMA broker, institutional traders will need a prime broker and retail traders can create algo trades with an MT4 broker. However, the main things to consider when looking for the best broker for algo trading are:

• Market access
• Pricing
• DMA
• API access
• Capitalisation and regulation

Types of algo trading strategies

There are many types of algo trading strategies, such as:

• Technical Indicators – most famous of them is ‘trend following’ strategies
• Mean reversion (or pair trading) – betting on similar securities moving together or apart
• Arbitrage – betting on similar securities have the same price, say between stock index futures and the underlying stocks
• Transactional – breaking orders to reduce costs such as VWAP (volume-weighted average price)
• High-Frequency – trading instruments multiple times within a second
• Machine learning – applying newer strategies such as deep learning to capture patterns

Of course, there is nothing to stop a fund from using a combination of all these strategies to gain an ‘edge’ over its fellow traders.

Some strategies, however, are more capital intensive than others. The initial investment can run into millions.

Factors to consider when setting up algo trading

1. Strategy – what type of trading strategy do you intend to employ? What is the target market? Do you believe you have an edge over the more mature funds? Some strategies are probably better suited for professional trading firms.
2. Data – what type of data do you need for your strategy? If you require tick-by-tick data, it would be quite expensive to buy. By contrast, end-of-day data is much cheaper.
3. Data Storage – where would you be storing your data? Buy or rent space?
4. Software – how would you be absorbing and analysing all the incoming data? Which programming language? Do you need to hire quality programmers to build a trading system? How reliable is the software?
5. Brokers – who would you be using to execute your trades? How much capital do they need to run an algo trading model?
6. Risk management – on top of buy/sell signals, you will need a robust risk management system to allocate positions, calculate weighting, and monitoring open positions. For example, how do you deal with drawdowns and volatile periods?
7. Research – do you continually run backtesting on new models? Good quant ideas are rare. Firms that found them will exploit them to the hilt.
8. Initial capital – how much capital do you have to start trading? If the initial capital is too small, you may not be able to withstand prolonged periods of drawdown.

All these factors are important. It takes time to think about all these input and a lot of effort to make sure they integrate in one coherent system.

More importantly, how involved would you be in setting up the business? Does it suit your personality, skillset and temperament? Remember, even with all these systems set up and running, the return on capital may be too volatile or too low. There is no guarantee it will hit your target returns.

Example – Setting up an automated trading strategy

One of the simplest examples of a rule-based strategy is also an old one: buy and sell a stock based on moving averages.

A moving average is a smoothing calculation to average past prices. The parameter N is the number of days to average. The larger the N, the smoother the average line. The chart below shows the 150-day moving average on British American Tobacco (BATS).

The rule: Buy stock when prices move above the moving average; sell stock when prices slip below the moving average; do nothing when prices is at the moving average.

At first glance, the rule is simple enough.

However, when we get to the actual implementation, things become more complicated. For instance,

• How much to buy and sell the stock as a percentage of total assets?
• Do we need additional filters such as waiting for a confirmed signal ie closing price above moving average?
• Do we add more (‘leverage up’) when the trade is going well?
• Is there a max loss limit (stop losses) on the trade?
• What about intra-day movements – do we ignore or factor into the system?
• How we trade the stock – at market or limit or near closing?

Furthermore, at the portfolio level how many stocks are we going to trade? Is there a particular sector focus?

Finally, is the parameter 150 good enough? Is there any other better possibilities such as 140 or 78? Have we tested the model on other stocks? Is the result consistent enough to warrant more capital allocation to this particular strategy?

As you can see, even when the trading rule is very simple, the design of the trading model is not. There are many other factors at work – and questions to answer – even before we bet a penny on a trading strategy. Many things can go haywire.

Once the strategy is mapped out, you will need to ‘test run’ it. That is, use real prices to generate p-and-l but without employing any real money. This is to assess the strategy in real-time.

Successful backtesting an algo trading strategy

Backtesting is a cornerstone of algo trading. This process assesses the profitability and reliability of trading models. It answers the question ‘should we trade it?’

What is backtesting? It is applying trading system to historical data. Backtesting allows us to visualise how well – or poor – the model is.

Some common yardsticks to measure a model’s overall characteristics:

• Profits and losses – total, max profits and loss, trade-by-trade pnl
• Volatility  – of each trade, drawdowns and total equity curve
• Returns – including risk-adjusted returns over time
• Distribution of trades – is the system dependent on a small number of large trades?
• Mechanics of trade – how volatile are the results based on different assumed entry prices, slippage and transaction costs

Once you have the system results, it gives you an idea on how well the model is working historically. However, there are some biases in these testing results, including:

• Optimisation bias – fitting the model to data which has no predictive value
• Look-ahead bias – an error that included future data for current calculation
• Survivorship bias – using data that are not representative of the whole sample

To mitigate these biases, you start with data. Are the data bias free? A good quality dataset is used preferred. Second, conduct out-of-sample statistics tests to examine the original results. Examine trading models with sensitivity tests. Simulating parameters or bootstrapping data and see if model results are good enough.

A good backtesting process takes time and effort to set up. It is crucial in giving traders confidence in their activities.

Where to learn algo trading for beginners?

If you intend to use algo to trade, there are many websites dedicated to this activity where you can learn, including:

• Quantpedia
• https://blog.quantinsti.com
• Quantstart
• Books, e.g.  Ernest Chan’s Algorithmic Trading, Robert Carver’s Systematic Trading
• Online courses such as udemy or coursera.org

Algo trading FAQs: