This paper tested two of the simplest and most popular trading rules – Auto-
Regressive Models and Moving Averages – by utilising the Australian Dollar relative to US Dollar from 1 Jan 1986 to 9 June 1999.

This data set was used by Tan [1995, 1997] in his study in comparing the profitability of systems based on Artificial Neural Networks and ARIMA models. Similar works were done earlier [LeBaron et al. 1995] which utilised Dow Jones index from 1897 to 1986. This paper did not utilise any index data due to the inconsistency of its composite stocks from time to time. The main reason for using the techniques was that they were simple to interpret and calculate, and seemed to work quite well in trending markets. Trading rules were derived from the short and long-term moving averages with the trading signals based on the differences between the two.

Combination of relatively longer period moving averages generally outperformed the shorter period moving averages. This was probably the result of eliminating unprofitable whipsaw trades. Buying (selling) signals were generated if the short (long) period moving average crossed above (below) the long (short) period moving average. Certain bands or filters were introduced to reduce the number of unnecessary trades that signals were only generated if the differences between the moving averages exceeded the interest rate differentials and foreign exchange spreads. Periods used were 5, 10, 15, 20 and 25 days for the short-term and 50 to 100 days for the long-term periods. Extensive tests to compare each and every moving average periods to find the best profit were carried out and the highest percentage of winning trades over the test period.

The work was extended to utilising support and resistance line as a filter to the buying or selling signals. Trading signals were generated only if the period tested was at the local minimum or maximum, or in other words, identifying key reversal areas. Results
confirmed that the use of two-period moving averages with auto-regressive models outperform the simple single-period moving averages. The use of support and resistance
lines as part of the filter rules will help a trading system to eliminate unnecessary trading, even though the overall performance does not outperform the previous two models. Keywords: Moving Averages, Short and Long Term Moving Averages, Auto-Regressive Models, Trading Systems, Foreign Exchange, Australian Dollar Market, Random Walk Theory, Technical Analysis, Support and Resistance, Trading Break-out Rules.

Forecasting foreign exchange rates or profiting from trading foreign exchange has been an extremely difficult task and most previous studies have shown little or no success in their attempts to predict foreign exchange market. Recently, this has been changing in both academic communities and financial industries. This paper presents the main features of one model or trading system being developed to generate profits out of trading foreign exchange. Traders considered these exchange rates to have persistent trends that permitted mechanical trading systems (systematic methods of repeatedly buying and selling on the basis of past prices and technical indicators) to consistently generate net profits with relative acceptable amount of countable risk. On the other hand, some other researchers presented evidence supporting the random walk hypothesis, which implies that rate changes are independent and have identical statistical distributions.

When prices follow random walk, the only relevant information in the historical series of prices, for traders, is the most recent price. The presence of a random walk in a currency market is a sufficient condition to the existence of a weak form of the efficient market hypothesis, i.e. that historical price movements could not be used to predict future prices. While there is no final word agreed between traders and academicians about the efficiency of the foreign exchange market, the old fashioned view in economic books that exchange rates follow a random walk has been dismissed by many research works [Tenti 1996]. There is however strong evidence indicating the returns are not independent of past changes. The term "Technical Analysis" is believed to be the original form of investment analysis [LeBaron 1995]. Technical Analysis attempts to forecast prices by studying the historical prices and a few related summary statistics about trading securities. From the Technical Analysis literature, works by LeBaron et al. [1995], provided strong support for the technical analysis being able to predict some variability on the financial markets.

They tested the two most popular trading rules – Moving Averages and Trading Range Break by utilising the Dow Jones Index from 1897 to 1986. The standard statistical analysis is extended through the use of bootstrap techniques and still found the techniques to be worth considering. The results of their research are consistent with the popular belief that technical rules have predictive power and outperform some other techniques. It shows that the rule-generating process of stocks is probably more complicated than suggested by the various other studies using linear models.

It is quite possible that technical rules are able to identify some of the patterns otherwise hidden. This seems to be the case as the authors emphasise that their successful systems were based on the simplest trading rules such as moving average techniques. However, other factors that need to be carefully considered were overlooked in their research. Transaction and brokerage costs should be included in the trading system calculation before they would be practically implemented. In this paper, similar test will be performed using foreign exchange data, since indices change their composite stocks from time to time, therefore distorting the forecasting

outcomes. Further more, in real life, one would be interested not only in efforts in forecasting but also in practical trading strategies with possibility of taking positions in the market. Tsoi, Tan and Lawrence [1993] in their earlier studies have shown that the direction of the forecast is more important than the actual forecast itself in determining the profitability of a model. Thus, the effort is always on to beat the market by superior techniques. For that reason, the work was further extended to build a trading system based on the rule-generating process over thirteen-year period. Again, this cannot go on forever as the market can ’learn’ and adapt to such techniques and strategies and can start following them. This confirms the economic theory of Efficient Market Hypothesis, which in its weakest form states that future prices cannot be predicted based on the past. One of the limitations to this test is that Technical Analysis or Time Series Analysis techniques do not include or take into account a number of factors such as macroeconomical or political effects, whether it be national or international, which may seriously influence the foreign currency market. Technical Analysis as its name suggests does not study the cause of the price move; it is the studies of the pattern of the price movements.

1.1 Data

In this exercise the time series data being used are as follows:

· Closing price of Australian Dollar quoted on weekly basis relative to US Dollar between 1st January 1986 and 23rd June, 1999, obtained from the Reserve Bank of Australia,

· The weekly Australian closing cash rate in Sydney from 1st January 1986 to 23rd June 1999 and obtained from the Reserve Bank of Australia,

· The weekly closing US Fed Fund rate in New York from 1st January 1986 to 23rd June 1999 and obtained from Federal Reserve Bank of Chicago, USA.

The optimum technical model is built using the in-sample data, starting from 1st January 1986 to 25th July 1998. The model is then tested on out-sample data from 2nd August 1998 to 23rd June 1999 and profits were calculated from the trades on those dates.

Next: Research Methodology and Results

Summary: Index