Predictable Responses in Currency Markets to Macroeconomic News: A Trading System Approach
Warwick Schneller
Abstract
This paper analyses how the release of a macro news event affects exchange rate behaviour. The event examined was the US non-farm payrolls announcement and the British Pound (GBP)/ US Dollar (USD) were the selected currency pair. A trading system model was developed based on a formal methodology previously applied to equity markets. The system examined the currencies reaction to the announcement in determining whether any behavioural patterns were present. Based on the trading system, no exploitable trading patterns were found.
Acknowledgement
The data for this project is supplied by Securities Industry Research Centre of Asia-Pacific (SIRCA) on behalf of Reuters.
Keywords
Foreign exchange, macroeconomic news, trading system
- Introduction
The foreign exchange market is by far the largest financial market in the world, with more than US$3 trillion of value traded daily across products in 2007, the most recent year for which global data is available(Bank of International Settlements, 2007) Despite its tremendous size and importance the question of explaining and predicting exchange rate movements is largely unanswered.
In this paper we examine how a macro news event (US employment) is impounded into a currency pair. Traditional asset market models of exchange rate determination, based on rational expectations and efficient markets, imply that announcements of public information are directly impounded in prices with there being no role for trades in this process of information assimilation (Love & Payne, 2009). Recent research has examined foreign exchange markets from a market microstructure perspective and found that “trading” is an important factor in the price formation process (Berger, Chaboud, Chernenko, Howorka, & Wright, 2008; Love & Payne, 2009) .
The general approach to examine how macro news events affect the behaviour of currencies has been through the application of traditional econometric models (Ehrmann & Fratzscher, 2005). The application of such methods does not necessarily capture ‘trading behaviours’ such as herd behaviour or overreactions.
This paper takes a different approach to examine the link between fundamentals and exchange rate movements. We apply a trading system model to examine the reaction of the British Pound versus the US Dollar (GBP/USD) to the release of United States employment data.
Although restricting our prediction of future currency prices on nothing more than a predictable event may seem too restrictive to be of any interest this can yield significant insights into the behaviour of asset prices. The study adopts a novel approach to examine whether intraday trading patterns exist surrounding the release of predictable information. This research has implications for the development of market microstructure and behavioural-based models.
The paper is organised as follows. Section 2 reviews foreign exchange market literature. Section 3 describes our data set and the structure of the foreign exchange market. Section 4 presents the papers methodology for examining the effects of US employment data on currency level and trading behaviour. Section 5 is a discussion of the results. Section 6 is future work and section 7 concludes.
- Literature Review
The seminal work of Meese & Rogoff (1983) found that especially at short horizons, a random walk forecast of exchange rates generally outperforms traditional predictive models drawn from economic theory; including purchasing power parity (PPP), uncovered interest rate parity (UIP), and monetary balance models of exchange rates. Numerous empirical studies since have investigated the time series behaviour of exchange rates and the empirical distribution of exchange rates. The result from these empirical studies is that the structural models for predicting exchange rates fail when tested out of sample (Berkowitz & Giorgianni, 2001; Faust & Rogers, 2003). At short time horizons there remains no well accepted “model” of exchange rate determination (Froot & Ramadorai, 2005).
This frustration has lead to a search for alternatives that better explain exchange rate changes. Ehrmann & Fratzscher (2005) argue that two approaches have emerged in the literature that have made progress in understanding exchange rate dynamics at short to medium term horizons. One of these approaches suggests that the chartist behaviour of market participants i.e. the pursuit of technical trading rules may account for some of the large movements and overshooting of currencies. An example of such research is Olser (2003) which examined the clustering of stop-loss and take-profit orders in the context of support and resistance levels. Further, Lo, Mamaysky, and Wang (2000) implemented an automated “charting” approach to identify trading patterns and found that certain methods had practical value. The paper utilises both findings to develop the trading system parameters.
The second and more recent approach is based on the seminal work of Evans and Lyons (2005b) which has shown that exchange rates at short-term horizons are to a significant extent driven by order flow, i.e. excess buying initiated or seller-initiated trading which reflects the market information processing mechanism. The finding that trading is an important factor in price formation differs to traditional asset market models of exchange rate determination. Under rational expectations and efficient markets, the announcements of public information are directly impounded in prices with there being no role for trades in this process of information assimilation (Love & Payne, 2009).
Evans and Lyons(2008) and Love and Payne (2009) find that much of this order flow is in fact closely linked to news about fundamentals. However, despite the role of order flow in the assimilation of public information into prices they do not suggest that foreign exchange markets are inefficient. They find that virtually all of the price changes associated with macro news announcements occur within the first two minutes of the release (Love & Payne, 2009).
Most of the literature on announcement effects has considered the effects of news releases on the level of asset prices, more specifically on the conditional mean of asset returns[1] (A. Chaboud et al., 2004) and generally applied an event study methodology. Andersen, Bollerslev, Diebold, & Vega (2003) found that the conditional mean of exchange rates adjust quickly to the release of news., effectively amounting to “jumps”. In contrast, the conditional variance adjustments were found to be more gradual, and that an announcement’s impact depends on its timing relative to other related announcements and on whether the announcement time is known in advance. Andersen et al (2003) examined the impact of employment data on an intraday basis and found the price effect was statistically significant.
Previous studies that have examined the effect of macroeconomic news announcements on foreign exchange markets have focussed on how anticipated information differs from the realised information. Kim, McKenzie, & Faff (2004) reason that market participants formulate expectations regarding upcoming information and therefore would only react to the unexpected component of information. Kim et al (2004) found that it is the “news” component which causes markets to react using daily data. This finding is supported by Andersen et al (2003) using an intraday event study method. An interesting finding by Chaboud, Chernenko, & Wright (2008) found that volume levels increase significantly on the release information even when in line with market expectations. Chaboud et al. (2008) used a high frequency data set which is likely to have contributed to the precision of the results.
This study adds to the previous findings by suggesting that the occurrence of an event is information in itself. For example, the publication of employment data in-line with average market expectations removes uncertainty and will cause associated trading activity which is consistent with the Chaboud et al (2008) finding.
The remainder of this section examines trading system research.
The use of automated trading systems has been actively researched by practitioners and academics but from differing perspectives. Practitioners have been driven by a profit maximisation motive. The Bank of International Settlements (2007) recognised the increased application of rule-based trading in currency markets. Statistics of the proportion of trading by autonomous trading systems in currency markets is not currently available. Academics have used them largely in the examination of market inefficiency and as a demonstration of machines ability to learn/replicate human functions.
We suggest that a trading system method provides an alternate way to measure currency behaviour to macroeconomic news. Econometric methods to date which model short-term exchange rate behaviour have performed poorly (Ehrmann & Fratzscher, 2005). Further, this static approach does not capture the complex behaviour of market participants, for example herd behaviour, over reactions, asset pricing bubbles (Ehrmann & Fratzscher, 2005).
- Description of FX market and the Data Set
The Foreign Exchange market is an over the counter (OTC) market. The absence of a centralised exchange creates a unique microstructure environment, dealing occurs via a decentralized multiple-dealer market in which three trading mechanisms operate simultaneously; direct dealer to dealer (interbank) transaction, broker dealt transactions and non-interbank customer dealt trades (Evans, 2002). Interbank dealers constitute 43 percent of the daily volume (Bank of International Settlements, 2007).
Currencies are dealt 24 hours a day in the interbank market, but are mostly inactive during weekends and national holidays. The trading week commences at 22:30 Greenwich Mean Time (GMT) which signals the opening of the Asian market and ceases at approximately 22:30 GMT on Friday which coincides with New York 5pm (Guillaume et al., 1997).
The data for this project is supplied by Securities Industry Research Centre of Asia-Pacific (SIRCA) on behalf of Reuters. The data sample consists of 15 minutely intraday spot[2] transactions, which were filtered for anomalies, e.g. out-of-range prices. Specifically the data is from the Reuters D2000-2 system. Thus our data contains no information on customer-dealer FX trades or on direct trades (non-intermediated) trades between dealers. Moreover, it should be noted that the trades occurring on D2000-2 should be regarded as public in the sense that they are visible to any party looking at a D2000-2 platform (Love & Payne, 2009).
The data set span is nearly 14 years, which is a much longer sample set than used in existing work that examines price responses to news. This allows an in-depth examination of price reaction to individual announcement, without having to aggregate them into aggregate measures (Evans & Lyons, 2005a).
The specific event for this study was the release of the United States Department of Labour nonfarm payrolls data and the currency selected was GBP/USD. The data set consisted of 173 non-farm employment releases. The timing of the release is advised to the market well in advance and is released to the public at 08:30 EDT[3].
The importance of individual macroeconomic variables shifts over time (Cheung & Chinn, 2001). The US non-farm employment report was selected as it was found to have a major impact on the prices of assets consistently (Andersen et al., 2003; Graham, Nikkinen, & Sahlström, 2003).
A criticism of certain studies examining the link between foreign exchange and macroeconomic variables, for example the Meese & Rogoff (1983) study was that forecasts were based on realized value in the forecast period (Faust & Rogers, 2003). The criticism is that by giving the model actual future data in forming the forecast, an artificial advantage is created. This does not affect this study since the specific numbers are not of interest only the occurrence of the event.
The currency pair selected for the study was the British Pound (GBP) and United States Dollar (USD). This currency pair represents one of the major trading currencies. This pair accounts for 12 percent of daily volume(Bank of International Settlements, 2007).
- Methodology
We apply a rule based trading system to examine the market microstructure of foreign exchange markets. This methodology differs from previous research which largely uses linear statistics and event study methods (Evans & Lyons, 2005a). The impetus for applying a trading system approach is to provide a different perspective and information on how a macroeconomic announcement affects the behaviour of a currency pair.
The application of a trading system approach gives consideration to ‘real world’ constraints such as; liquidity constraints, slippage on the execution of orders as well as risk management and money management and allows a practical interpretation of results.
A general criticism of using trading systems is the absence of a formal methodology describing the procedure for the development as well as the benchmarking of results (Vanstone & Finnie, 2009). To address this issue the paper adopted the empirical methodology developed by Vanstone and Finnie (2009) which has previously been applied to equity markets.
The general procedure is summarised below, and is discussed further in the following sections:
- Hypothesis development
- Partitioning of data
- System design
- In-sample testing and optimisation of parameters
- Out-of-sample testing
- Hypothesis
The null hypothesis in this study is that mechanical rules for generating trading signals should not result in unusual (risk adjusted) profits.
- Partitioning Available Data
The GBP/USD data set was separated into two data sets; in-sample and out-of-sample. The division of data into an in-sample and out-of sample set creates the opportunity to test the developed system on ‘unseen’ data and improve the validity of the results. A general criticism of research examining trading patterns is that the results are the product of data mining and cannot be replicated when applied to out-of-sample data (Jegadeesh & Titman, 2001).
Thawornwong & Enke (2004) document that the relationship between security prices and the variables that determine that price change over time. The change in the structural mechanics of markets implies that it is necessary for the data set to be separated ‘vertically’ rather than ‘horizontally’. The horizontal approach to partitioning splits the entire data set into either training or testing sets but not in chronological order. Vanstone & Finnie (2009) propose that the horizontal method is invalid when it is recognised that the system may know pricing movements and information which it could not have known in chronological time. This may lead to higher quality predictions without a valid basis. To avoid the potential for look-ahead bias[4] the data was divided vertically, creating two data sets; training set (in-sample) and testing set (out-of-sample) divided chronologically.
As a result of dividing the GBP/USD into two data sets it is necessary to determine an appropriate ratio for the split[5]. Pardo (1992) and Azoff (1994) suggest that the training period should be long enough to capture a variety of different market phases. Pardo (1992) also suggest that the longer the testing phase the greater the longevity of the model. Given the aforementioned, the study divided the data set as follows; in-sample (training): January 1996 till December 2003 and out-of-Sample (Testing): January 2004 till June 2010. This represents an approximate split of 60 percent training and 40 percent testing. The training period includes a number of different market phases (bull, bear and sideways markets) as well as ‘outlier’ (US terrorist attack type events). The out-of-sample (testing data set) also captures a variety of market conditions including the British terrorist attack and recent financial crisis.
- System Design
A rule based trading system was developed to determine whether abnormal profits could be earned using a predictable event. The system does not consider the actual value of the event or deviations from market estimates but simply the event itself. The system consisted of the following general parameters as advocated by Chande (2001) to provide a multilayered approach to system development.
The general system parameters are determined and then the specific values are set via an optimization process, discussed in section 2 D.
The following parameters were adopted;
| Entry & Exit Rules |
|
| Risk Management[6] |
|
| Money Management [7] |
|
Table 1: Trading sytem parameters
- In-Sample Bench Marking (Optimization)
The in-sample bench marking process was conducted to determine the relationship between trading system parameters and trading system performance. The broad rules of the system remain unchanged. For example, how does changing the level of the fixed stop affect performance?
In-sample bench marking was conducted using two approaches; optimisation and in-sample simulations. The optimization process generates a series of metrics for the system based on a set of inputs. The system then changes the value of the inputs and repeats the process. Optimization provides information on the interaction between parameter values and performance.
In-sample simulations provide a detailed set of metrics which allow examination of complete systems.
The following section discusses the output from the optimization.
Figure 1: Relationship between fixed stop level and profit target.
Figure 1 represents the interaction between the systems fixed stop (Optvar1) and profit target (Optvar2) and the systems overall profitability keeping all other variables constant.
The relationship between the fixed stop parameter and profit target can be interpreted as this systems willingness to take risk in return for potential profit. It was found that generally a positive relationship held between the fixed stop level and profit target.
Figure 2: Maximum adverse excursion distribution
The general relationship observed between the stop loss parameters, profit targets and overall system performance can be better understood when the actual ‘behaviour’ of trades are considered.
Figure 2 is the maximum adverse excursion distribution (MAE). The MAE is generated from a set of trades and determines the extent to which favourable (profitable) trades range into unprofitable territory prior to closing out profitably (Vanstone & Finnie, 2009). In figure 2 it is observed that the greater the loss incurred by a trade the lower the chance of returning to profitability. For example, in the in-sample data a loss of 5.00 percent of value resulted in 100.00 percent recovery versus a loss of 30.00 percent value which resulted in 100.00 percent of those positions being closed out for a loss.
Figure 3: Maximum favourable excursion distribution
The MAE distribution focuses on negative outcomes. Its opposite is the maximum favourable excursion (MFE) distribution. The MFE distribution provides two set of information. One set shows the largest profit that a trade incurred, while the second distribution shows the percentage of those trades that were profitable that subsequently became losses. For example in figure 3, five trades’ generated 30.00 percent profit but 60.00 percent of them eventually became losses.
The following section discusses the output from the in-sample simulations.
Four simulations were conducted on the in-sample data set. The general rules of each system were identical; each contained a fixed stop, trailing stop and profit target. Each simulation incrementally increased the risk tolerance and potential reward.
The parameter values of the in-sample simulations are as follows:
| Fixed stop | Trailing stop | Profit Target | |
| System 1 | 10 | 20,20 | 20 |
| System 2 | 50 | 40,40 | 40 |
| System 3 | 70 | 50,50 | 50 |
| System 4 | 90 | 60,60 | 60 |
Table 2: In-sample simulation parameters
| Simulation 1 | Simulation 2 | ||||||||
|
|
||||||||
| Simulation 1 | Simulation 2 | ||||||||
| Simulation 3 | Simulation 4 | ||||||||
Table 4: Profit distributions for in-sample simulations
Table 3 reports the performance of the simulations over the in-sample data period (January 2006 – December 2003)[8]. The table illustrates two key points. Firstly, changing the values of the system parameters has a significant impact on the overall performance of the system, despite the core basis of components of the system remaining unchanged. Secondly, a risk-averse set of parameters does not necessarily imply the lowest loss. System 1 was the most risk-averse of the system tested and it occurred the highest loss (-$1,290.46) as compared to system 2, 3 and 4 which resulted in profitable outcomes. The loss achieved by system 1 indicates that despite potentially profitable trading opportunities, as shown by systems 2, 3 and 4, the restrictive nature of the parameters prevents profits from being captured. Based on table 3 system 4 had the highest performance over the in-sample data set.
Table 4 reports the profit distribution for the in-sample simulations. The key reason for the profit distribution is to provide information on the nature of the trades that make up the overall performance of the system. This adds additional information as compared to table 3 which shows overall performance over time. For example, the reason for the overall loss incurred by System 1 (see table 3 and appendix 1) becomes is evident from table 4 which shows that losses were too frequent (72.00 percent) and too great (average loss 8.00 percent) relative to favourable trades to result in a profitable system. In terms of overall performance based on table 4 system 4 had the highest success.
From a trading system perspective system; 2, 3 and 4 were able to generate positive in-sample profits. This finding indicates that a potential exploitable trading pattern exist in GBP/USD data set following the release of US nonfarm employment data. Based on the results of the optimisation and in-sample simulations trading system 4 was applied to the out-of-sample data set.
- Empirical Findings (Out-of-sample)
The developed trading system was applied to the out-of-sample data set (Jan 2004 – June 2010).
It was found that the system when tested on unseen data generate a negative result (-$1,020.32). Therefore, the alternate hypothesis is rejected. Table 3 reports the out-of-sample results. The first observation is that the system is unable to replicate a positive return as per the in-sample simulation. The results indicate the that ratio of winning trades to losing decreased when tested out of sample (56.00 percent out-of-sample versus 80.00 percent in-sample).
| Long + Short | |
| Net Profit |
-$1,020.32 |
| Profit per Bar |
-$3.37 |
| Number of Trades |
43 |
| Avg Profit/Loss |
-$23.73 |
| Avg Profit/Loss % |
-0.01% |
| Avg Bars Held |
7.05 |
| Winning Trades |
24 |
| Winning % |
55.81% |
| Gross Profit |
$8,949.84 |
| Avg Profit |
$372.91 |
| Avg Profit % |
0.21% |
| Avg Bars Held |
5.71 |
| Max Consecutive |
6 |
| Losing Trades |
19 |
| Losing % |
44.19% |
| Gross Loss |
-$9,970.16 |
| Avg Loss |
-$524.75 |
| Avg Loss % |
-0.30% |
| Avg Bars Held |
8.74 |
| Max Consecutive |
4 |
| Max Drawdown |
-$2,630.15 |
| Max Drawdown Date |
4/12/2009 |
Table 3: Out-of-sample performance
Figure 4 are the systems results across time. What was observed is that the system was unprofitable for the majority of the tested period (Jan 2004 – Aug 2009). This differs from the in-sample result which found the system was able to generate positive returns over the entire data set.Figure 4: Profit curve (out-of-sample)
Figure 5 reports the breakdown of the trade. What is found provides an explanation for the different outcome between the in-sample and out-of-sample data sets. A majority of the trades were profitable (58.81 percent). Despite the higher frequency of favourable outcomes the greater weight of the losses ultimately leads to the diminished performance. The average loss for the system was -$542.75 as compared to the average profit $372.91. This is reflected in figure 5 with the ‘fatter’ left tail.
Figure 5: Profit distribution (out-of-sample)
- Future Work
The first possible avenue for future research is to examine whether the trading performance of this system can be improved by feeding other information than price data into the system. In earlier work (Love & Payne, 2009) demonstrated that order book or order flow information is able to enhance the predictive performance. Additionally, future studies will investigate a wider range of economic data releases across additional currency pairs.
- Conclusions
In this paper we developed an automated trading system based on the publication of US Department of Labour non-farm payrolls data. The study was undertaken to contribute to the literature examining market microstructure in foreign exchange markets and how market participants impound information into prices.
Traditional asset market models of exchange rates, based on rational expectations and efficient markets, imply that public announcements are directly impounded into prices with no role for trading. Recent research has found that ‘trading’ is an important factor in the price formation process and that the behaviour of financial agents is not always consistent with rational expectations.
We developed a trading system that was developed on the basis of a formal methodology previously applied to equity markets. The paper departs from previous approaches that have applied a linear econometric approach to provide a more applied interpretation of the results.
The system examined whether a rule based trading system based on a predictable event could generate profits. The currency pair examined was the GBP/USD and the specific event was the US non-farm payrolls, which is published monthly. The study examined intraday data from 1996-2010.
The system attempted to predict the reaction of market participants following an event not the actual announcement itself. It was found that the developed system was unable to generate a positive return on out-of sample data. The results indicates the following; Based on the system parameters used in this paper there are no predictable behaviour or exploitable trading patterns following US nonfarm payrolls data in the GBP/USD currency pair.
References
Andersen, T. G., Bollerslev, T., Diebold, F. X., & Vega, C. (2003). Micro effects of macro announcements: Real-time price discovery in foreign exchange. American Economic Review, 93(1), 38-62.
Azoff, E. M. (1994). Neural network time series forecasting of financial markets: John Wiley & Sons, Inc. New York, NY, USA.
Balsara, N. J. (1992). Money management strategies for futures traders: John Wiley & Sons Inc.
Bank of International Settlements. (2007). Triennial Central Bank Survey of Foreign Exchange and Derivatives Markets Activity: Bank of International Settlements.
Berger, D. W., Chaboud, A. P., Chernenko, S. V., Howorka, E., & Wright, J. H. (2008). Order flow and exchange rate dynamics in electronic brokerage system data. Journal of International Economics, 75(1), 93-109.
Berkowitz, J., & Giorgianni, L. (2001). Long-horizon exchange rate predictability? Review of Economics and Statistics, 83(1), 81-91.
CFA Institute. (2010). CFA(R) Program Curriculum : Equity (Vol. 4): Pearson.
Chaboud, A., Chernenko, S., Howorka, E., Iyer, R. S., Liu, D., & Wright, J. H. (2004). The High-Frequency Effects of U.S. Macroeconomic Data Releases on Prices and Trading Activity in the Global Interdealer Foreign Exchange Market. SSRN eLibrary.
Chaboud, A. P., Chernenko, S. V., & Wright, J. H. (2008). Trading activity and macroeconomic announcements in high-frequency exchange rate data. Journal of the European Economic Association, 6(2-3), 589-596.
Chande, T. S. (2001). Beyond technical analysis : how to develop and implement a winning trading system (2nd ed.). New York: Wiley.
Chande, T. S. (2001). Beyond Technical Analysis: how to develop and implement a winning trading system: John Wiley & Sons Inc.
Cheung, Y. W., & Chinn, M. D. (2001). Currency traders and exchange rate dynamics: a survey of the US market. Journal of International Money and Finance, 20(4), 439-471.
Ehrmann, M., & Fratzscher, M. (2005). Exchange rates and fundamentals: new evidence from real-time data. Journal of International Money and Finance, 24(2), 317-341.
Evans, M. D. D. (2002). FX trading and exchange rate dynamics. The Journal of Finance, 57(6), 2405-2447.
Evans, M. D. D., & Lyons, R. K. (2005a). Do currency markets absorb news quickly? Journal of International Money and Finance, 24(2), 197-217.
Evans, M. D. D., & Lyons, R. K. (2005b). Meese-Rogoff redux: Micro-based exchange-rate forecasting. American Economic Review, 95(2), 405-414.
Evans, M. D. D., & Lyons, R. K. (2008). How is macro news transmitted to exchange rates? Journal of Financial Economics, 88(1), 26-50.
Faust, J., & Rogers, J. H. (2003). Exchange rate forecasting: the errors we’ve really made. Journal of International Economics, 60(1), 35-59.
Froot, K. A., & Ramadorai, T. (2005). Currency returns, intrinsic value, and institutional-investor flows. The Journal of Finance, 60(3), 1535-1566.
Graham, M., Nikkinen, J., & Sahlström, P. (2003). Relative importance of scheduled macroeconomic news for stock market investors. Journal of Economics and Finance, 27(2), 153-165.
Guillaume, D. M., Dacorogna, M. M., Davé, R. R., Müller, U. A., Olsen, R. B., & Pictet, O. V. (1997). From the bird’s eye to the microscope: A survey of new stylized facts of the intra-daily foreign exchange markets. Finance and stochastics, 1(2), 95-129.
Jegadeesh, N., & Titman, S. (2001). Profitability of Momentum Strategies: An Evaluation of Alternative Explanations. The Journal of Finance, 56(2), 699-720.
Kim, S., McKenzie, M., & Faff, R. (2004). Macroeconomic News Announcements and the Role of Expectations: Evidence for US Bond. Stock and Foreign Exchange MarketsV Journal of Multinational Financial Management, 14.
Lo, A. W., Mamaysky, H., & Wang, J. (2000). Foundations of Technical Analysis: Computational Algorithms, Statistical Inference, and. Journal of Finance, 55(4), 1705-1765.
Love, R., & Payne, R. (2009). Macroeconomic news, order flows, and exchange rates. Journal of Financial and Quantitative Analysis, 43(02), 467-488.
Meese, R. A., & Rogoff, K. (1983). Empirical exchange rate models of the seventies:: Do they fit out of sample? Journal of international economics, 14(1-2), 3-24.
Osler, C. L. (2003). Currency orders and exchange rate dynamics: an explanation for the predictive success of technical analysis. The Journal of Finance, 58(5), 1791-1820.
Pardo, R. (1992). Design, testing, and optimization of trading systems: John Wiley & Sons Inc.
Thawornwong, S., & Enke, D. (2004). The adaptive selection of financial and economic variables for use with artificial neural networks. Neurocomputing, 56, 205-232.
Vanstone, B., & Finnie, G. (2009). An empirical methodology for developing stockmarket trading systems using artificial neural networks. Expert Systems with Applications, 36(3, Part 2), 6668-6680.
Appendix 1: In-sample simulations
| Fixed Stop Loss: 10, Profit Target: 20, Trailing Stop: 20,20 | Fixed Stop Loss: 50, Profit Target: 40, Trailing Stop: 40,40 | ||
| Long + Short | Long + Short | ||
| Net Profit |
-$1,290.46 |
Net Profit |
$4,879.53 |
| Profit per Bar |
-$14.83 |
Profit per Bar |
$2.64 |
| Number of Trades |
50 |
Number of Trades |
50 |
| Avg Profit/Loss |
-$25.81 |
Avg Profit/Loss |
$97.59 |
| Avg Profit/Loss % |
-0.02% |
Avg Profit/Loss % |
0.06% |
| Avg Bars Held |
1.74 |
Avg Bars Held |
37.02 |
| Winning Trades |
14 |
Winning Trades |
32 |
| Winning % |
28.00% |
Winning % |
64.00% |
| Gross Profit |
$3,049.91 |
Gross Profit |
$13,969.71 |
| Avg Profit |
$217.85 |
Avg Profit |
$436.55 |
| Avg Profit % |
0.14% |
Avg Profit % |
0.28% |
| Avg Bars Held |
1.5 |
Avg Bars Held |
40.44 |
| Max Consecutive |
2 |
Max Consecutive |
10 |
| Losing Trades |
36 |
Losing Trades |
18 |
| Losing % |
72.00% |
Losing % |
36.00% |
| Gross Loss |
-$4,340.37 |
Gross Loss |
-$9,090.17 |
| Avg Loss |
-$120.57 |
Avg Loss |
-$505.01 |
| Avg Loss % |
-0.08% |
Avg Loss % |
-0.32% |
| Avg Bars Held |
1.83 |
Avg Bars Held |
30.94 |
| Max Consecutive |
11 |
Max Consecutive |
6 |
| Max Drawdown |
-$2,100.31 |
Max Drawdown |
-$4,310.14 |
| Max Drawdown Date |
7/09/2001 |
Max Drawdown Date |
1/09/2000 |
| Fixed Stop Loss: 70, Profit Target: 50, Trailing Stop: 50,50 | Fixed Stop Loss: 90, Profit Target: 60, Trailing Stop: 60,60 | ||
| Long + Short | Long + Short | ||
| Net Profit |
$12,999.64 |
Net Profit |
$16,369.63 |
| Profit per Bar |
$3.86 |
Profit per Bar |
$3.76 |
| Number of Trades |
50 |
Number of Trades |
50 |
| Avg Profit/Loss |
$259.99 |
Avg Profit/Loss |
$327.39 |
| Avg Profit/Loss % |
0.17% |
Avg Profit/Loss % |
0.21% |
| Avg Bars Held |
67.34 |
Avg Bars Held |
87.02 |
| Winning Trades |
39 |
Winning Trades |
40 |
| Winning % |
78.00% |
Winning % |
80.00% |
| Gross Profit |
$20,739.77 |
Gross Profit |
$25,399.73 |
| Avg Profit |
$531.79 |
Avg Profit |
$634.99 |
| Avg Profit % |
0.34% |
Avg Profit % |
0.41% |
| Avg Bars Held |
62.1 |
Avg Bars Held |
80.5 |
| Max Consecutive |
10 |
Max Consecutive |
18 |
| Losing Trades |
11 |
Losing Trades |
10 |
| Losing % |
22.00% |
Losing % |
20.00% |
| Gross Loss |
-$7,740.13 |
Gross Loss |
-$9,030.10 |
| Avg Loss |
-$703.65 |
Avg Loss |
-$903.01 |
| Avg Loss % |
-0.45% |
Avg Loss % |
-0.57% |
| Avg Bars Held |
85.91 |
Avg Bars Held |
113.1 |
| Max Consecutive |
2 |
Max Consecutive |
2 |
| Max Drawdown |
-$1,790.02 |
Max Drawdown |
-$2,750.02 |
| Max Drawdown Date |
3/10/1997 |
Max Drawdown Date |
4/12/1998 |
[1] The conditional mean and conditional variance of exchange rate returns refer to the expectation and variance of exchange rate returns given all information up to the time of the announcement, including the announcement itself (A. Chaboud et al., 2004).
[2] single outright transaction involving the exchange of two currencies at a rate agreed on the date of the contract for value or delivery (cash settlement) within two business days (Bank of International Settlements, 2007).
[3] Except in instances of mistaken release. The estimate for November 1998 (employment data for October 1998) was scheduled to be released at 8.30am EST on Friday, November 6th. Some estimates for nonfarm payroll employment inadvertently were released prematurely on the internet, so the schedule was revised and the report was officially released at 1.30pm EST on Thursday, November 5th.
[4] Look-ahead bias is the use of information that was not contemporaneously available at the time of decision making (CFA Institute, 2010).
[5] Vanstone & Finnie (2009) provide a detailed review of the literature of splitting data.
[6] In a trading context risk as defined by Balsara (1992) and Chande (2001) is the ‘risk of ruin’ and in a broader context the downside risk of a trade. Trading risk is generally managed through the implementation of stop loss orders.
[7] Money management relates to how much risk a decision maker should take relative to the expected reward. Money management in an economics context is what percentage of wealth should be risked in order to maximize a decision maker’s utility function (Balsara, 1992).
[8] Appendix 1 are the detailed simulation results.
