Chapter 4 Foundations of Strategy Development

This chapter will focus on the fundamentals of strategy development, including.

  • An introduction to common strategy types.
  • Common Strategy Evaluation Indicators.
  • The strategy development process.
  • Frequently asked questions about strategy development.
  • Using TqSdk for strategy development and backtesting.

4.1 Introduction to Common Strategy Types

A quantitative trading strategy is a set of objective “trading rules”, which are the conditions that must be met in order to buy or sell an order. In this subsection, we will review the existing research on trading strategies.
In general, trading strategies can be categorized into two types: the first type of strategy is to first build a model that predicts the direction of market price or trend changes, and then trade based on that prediction. This type of strategy usually uses machine learning techniques to predict the direction of market prices or trends. Next decisions are made as to when to buy or sell orders.
The second category of strategies consists of trading strategies that do not rely on any predictive modeling, such as technical indicators, momentum, arbitrage, etc.

4.1.1 Forecast-based

In 2009, Li et al. proposed a framework for predicting trend turning points. The model combines chaotic dynamics analysis with Artificial Neural Network (ANN) modeling. The goal is to capture the nonlinear and chaotic behavior of financial markets to predict potential turning points, and then add a Genetic Algorithm (GA) module to optimize predefined trading parameters to maximize the profit generated by the proposed trading strategy.

As a validation, they applied the trading strategy to the Dow Jones Industrial Average (DJIA) index time series and TESCO stock (UK). The experimental results show that applying the trading strategy to TFSCO stock (UK) can generate 69.78% annual return.

In 2012, Huang et al. proposed a methodology for population selection using support vector regression (support vector regression, support vector regression and genetic algorithms (Genetic Algorithms), where they used support vector regression models to predict and classify the profitability of stocks. The classification process involves the use of basic stock metrics (e.g., stock metrics, growth, profitability, liquidity), and the stocks that are categorized as “most profitable” are then used to form a
Portfolio. On the basis of the model, a genetic algorithm (GA) is used to optimize the parameters of the trading model. The experiments reported consisted of creating portfolios using 30 stocks. The results of the experiments showed that in the best case scenario, the trading system could generate an annual return of 17.57%.

In 2013, Evans et al. introduced prediction and decision-making models based on artificial neural networks (ANN) and genetic algorithms (GA) to predict changes in the direction of market trends. The data package used in this study consisted of historical exchange rates of GBP/USD, EUR/GBP and EUR/USD currency pairs for a length of 70 weeks. They reported that the trading strategy could generate an annualized return of 23.3%.In 2015, Giacomel et al. proposed an artificial neural network model (ANN) to predict the direction of price movements. They actually proposed two artificial neural network models:the first one predicts the opening and closing prices of a stock for the next period; while the second one is trained to predict the direction of the price trend for the next period. These two ANN models were combined to form a trading strategy. They tested the proposed model using 18 stocks selected from the North American and Brazilian stock markets. The experimental results show that the trading strategy can generate annualized returns of up to 76%.

In 2016, Chourmouziadis and Chatzoglou proposed a fuzzy trading system. They used a mixture of 4 technical indicators to predict stock prices, two of which are rarely used in research papers, the parabolic SAR and GANN-HiLo. Based on these 4 technical indicators, they proposed a total of 16 fuzzy rules. The fuzzy system assigns weights to each rule based on profitability during the training set. The backtesting is carried out using daily data from the Athens Stock Exchange for a period of more than 15 years, which is divided into bull and bear market periods. The experimental results show that the system generates fewer losses during bear market periods and smaller gains during bull market periods than the Buy and Hold strategy.

In 2016, Chen et al. proposed an intelligent pattern recognition model to predict turning points in the uptrend of stock prices (i.e., upside turning points). Their model used nine technical indicators as pattern recognition factors to identify stock patterns, and they employed rough set theory and genetic algorithms to predict upward turning points. Next, the authors build a trading strategy based on the proposed predictive model. In model validation, they evaluated the proposed model in two stock databases (TAIEX and NASDAQ). They report that the trading strategy generates, on average, a 57% annualized return.
In 2016, Gocken et al. proposed a model for predicting stock prices on the Istanbul Stock Exchange The proposed model uses a hybrid artificial neural network, where the inputs are technical indicators selected through a model combining the Harmony Search algorithm (HS) and the Genetic Algorithm (GA). They built a trading strategy based on the proposed predictive model and applied this trading strategy to the Turkish stock index BIST 100.They reported a return of 6.04% over 160 trading days.

Finally, we should note that although financial time series forecasting is a very attractive goal many studies did not build trading strategies based on their forecasting models. In fact, it is very important to establish a trading strategy that demonstrates that the proposed forecasting methodology can be profitable in the real market; after all, forecasting is not the same as profitability.

4.1.2 Trading strategies not based on predictive models

1. Technical trading

The first type of trading we consider is “technical trading”. Usually, technical traders analyze price charts in order to speculate on the possible direction of the market. This analysis employs a number of technical indicators, which look at past patterns to predict the future price direction of a security. Often the discovery of such patterns can help establish a delivery strategy (i.e., buy and sell rules). Examples of traditional technical indicators include: moving averages, Relative Strength Index Bollinger Bands, etc. The development of trading strategies based on technical indicators is very common in the academic world. This section will outline some of the technical trading strategies.

In 2009, Watson established a new methodology for studying the profitability of two technical indicators. He applied his methodology to daily data from January 1, 1980 to December 31, 2003 for 983 stocks traded on the London Stock Exchange. He concluded that the head-and-shoulders pattern had an average excess return of 5.5% per year.In 2009, Schulmeister' examined the profitability of 2,580 Technology Transaction Rules’ (TTRs). He reported that the profitability of these TTRs had declined steadily since 1960 and had been unprofitable since the early 1990s. However, based on 30-minute data, the same TTRs have averaged an annual return of 7.2%. For intraday trading, he reports, technical indicator trading is nonetheless entirely profitable.

In 2015, Cervell6-Royo et al. proposed a technical trading rule based on risk adjustment. They proposed a modified version of a technical indicator called “Flag Patterns” with the aim of “enhancing the robustness of flag patterns and their use in trading rule design”. They generated 96 different trading rules and applied them to three indices: the US Dow Jones Index (DJIA), the German DAX Index and the UK FTSE Index. The experimental results show that the trading rules were able to generate returns from November 26, 2004 to February 27, 2007 as high as 94.9%.

2. Momentum Strategy

The second type of trading is the “momentum strategy”, which does not depend on any trading pattern. In general, momentum strategies buy assets with higher recent returns and sell assets with lower recent returns.

In 2011, a study by the Bank for International Settlements (BIS) Monetary and Economic Department conducted an extensive empirical investigation into the profitability of momentum strategies in the foreign exchange market. The authors found that momentum strategies significantly favor minor currencies (i.e., currencies that are not actively traded in the FX market) with relatively high transaction costs. They also suggest that momentum strategies may lead to higher returns in the FX market than in the equity market.

In 2013, Daryl et al. proposed a momentum strategy based on a new risk-return criterion combined with an asset selection methodology. They attempted to create portfolios based on the introduced risk-return criteria by applying the model to the Korean stock market index (KOSPI 200) for the period from June 2006 to June 2012.The proposed momentum strategy was based on the risk-return criteria of the Korean stock market index (KOSPI 200), which is the largest stock market in the world. According to the report, the proposed momentum strategy did generate attractive positive returns.

3. Carry trade

In 2011, Bertolini studied the profitability of several Carry portfolio strategies (i.e. term structure strategies). He analyzed whether different asset allocations, market timing, and money management methods have the potential to improve the performance of simple Carry portfolios. The experiments were conducted using a dataset of G10 currencies for the period from January 1, 1999 to March 5, 2010.1 He considered the performance of various FX Carry portfolios. He considered various FX Carry portfolio strategies and found that the best strategy performance (e.g., 1-week returns) could be achieved by ranking currencies based on shorter duration returns In 2014, Laborda et al. proposed an asset allocation strategy aimed at improving the performance of currency Carry trades in which currencies were selected from the G10 currencies. Their model dynamically assigns weights to long and short positions in a portfolio. These weights are determined by financial variables that reflect changes in macroeconomic conditions and the likelihood of intertemporal collapse risk. They report that between January 2009 and February 2012, this asset allocation strategy generated significantly higher returns than the original Currency Carry trade.