back start next
[start] [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25] [26] [27] [28] [29] [30] [31] [32] [33] [34] [35] [36] [37] [38] [39] [40] [41] [42] [43] [44] [45] [46] [47] [48] [49] [50] [51] [52] [53] [54] [55] [56] [57] [58] [59] [60] [61] [62] [63] [64] [65] [66] [67] [68] [69] [70] [71] [72] [73] [74] [75] [76] [77] [78] [79] [80] [81] [82] [83] [84] [85] [86] [87] [88] [89] [90] [91] [92] [93] [94] [95] [96] [97] [98] [99] [100] [101] [102] [103] [104] [105] [106] [107] [108] [109] [110] [111] [112] [113] [114] [115] [116] [117] [118] [119] [ 120 ] [121] [122] [123] [124] [125] [126] [127] [128] [129] [130] [131] [132] [133] [134] [135] [136] [137] [138] [139] [140] [141] [142] [143] [144] [145] [146] [147] [148] [149] [150] [151] [152] [153] [154] [155] [156] [157] [158] [159] [160] [161] [162] [163] [164] [165] [166] [167] [168] [169] [170] [171] [172] [173] [174] [175] [176] [177] [178] [179] [180] [181] [182] [183] [184] [185] [186] [187] [188] [189] [190] [191] [192] [193] [194] [195] [196] [197] [198] [199] [200] [201] [202] [203] [204] [205]
120 The methods pre\iously discussed were based on pattems familiar to traders. The weekly and weekend studies, as well as the intraday time pattems shown in the pre™us sections, were originally verified by hand and later recalculated using a computer. There is a tjpe of pattem recognition, however, that would hardly be considered without the a\ailability of a computer. Rather than the conventional price pattems in which recurring sequences of higher and lower dajs are foimd within certain qualified intervals, computer-based pattem recognition refers to sets of descriptors and classes of interest. For example, Aronson describes the sets and values that must be satisfied by a professional jockey as: Descriptor Vdue-Rangc Height Under SS" Weight Under 120 lbs. Age 16 to 35 Years riding horses Over 10 The set of people who satisfy all four conditions are said to contain all professional jockejs. The converse, that all people satising these conditions are professional jockejs, is not true. It will be necessarj to qualify this set further to create a set that contains only professional jocke5s; however, these four conditions go a long way toward reducing the field How can this tool be applied toward the de\elopment of a trading strategj? if the sjstem is defined in terms of the trade profile, it becomes obvious. Consider the following characteristics of a trade: 1. The price moves higher or lower by at least 3"o of the starting price. 2. The price move occurs within 20 dajs. 3. There is no loss exceeding .5"o. Either a computer or an analjst can locate all price moves that satisfy these conditions within a price senes. Each of the 5 dajs preceding an upward move, which satisfies these conditions, can be marked as buy dajs, and the 5 dajs preceding a downward movement can be designated sell dajs (see Figure 15-9). The computer now contains the set of all buy and sell dajs-those days on which it would be good to get a buy or sell signal. Next, some likely indicators must be specified to be used for identiing that these trades are about to bin. The following might be used: 1. The mo\ing average direction 2. An overboughf oversold indicator such as a relative strength indicator (RSI) or contrarj opinion .5 David R. Aronson, Artificial Intelligence Methods (privately published). Also see David R. Aronson, "Artificial InteUigence/Pattem Recognition Applied to Forecasting Financial Market Trends," Market Technicians Association joumal (May 1985, pp. 91-131). FIGURE 15-9 Specific bujing and selling dajs.
Buyingdays Price mowes 3. The direction of changes in trading voliune 4. A 10-day momentum By entering a broad selection of indicators and trjing to avoid duplication, the computer can find imique values for combinations of indicators that primarily occur during the days selected as buy and sell periods. Ideally, all buy signals should occur when one indicator, or the value of combining indicators, exceeds a specific value. For example, all buy signals occur when the average value of the RSI and the maiket sentiment (contrarj opinion) is below the I0"o level. However, ha\ing all buy signals occur here is not enough. Poor signals may appear at this level, which cause large losses The perfect sjstem will have no losing signals occur in this zone. Unfortunately, in the real world there are no perfect solutions. The trades that are signaled by the combinations of indicators will have to be studied for net return, risk, and other performance criteria However, the technique of setting up classes of indicators, buy and sell dajs, is a new and valid approach to sjstem de\eIopment It is analogous to the multiple regression method used by econometricians to find the relationship between statistics and prices. Although the econometricians use inflation, supply, interest rates, and so forth, pattem recognition can employ technical indicators and discrete pattems to forecast a buy or sell day. ARTIFICIAL INTELLIGENCE METHODS Artificial mteUigence refers to a computer process that performs an operation corresponding to, or approaching, human thinking. This is intended to distinguish it fiom simple pattem recognition, with which il is often confused. The state of the art in artificial intelligence is the separation of two ideas. The collection of information that is stored in the brain has been termed the knowledge base. This is distinguished fiom reason, rules, and logic, called the inference engine. These ideas are not very different fiom the database and trading strategies that are discussed here The closest practical approach to artificial intelligence is heuristic programming. This refers to computer learning in very much the same way as finding the way out of a mazeThe computer starts with rules relevant to the problem, then records the successful and unsuccessful experiences. Eventually, it has a complete table of what to do for each situation, or at least a table of probable solutions. This is a realistic, intelligent approach when the same events can be expected to recur in the same way. It does not help in new situations without the added complication of incorporating extrapolation, basic relationships (e.g., price level to volatility), and other forms of expectations. The danger of the heuristic approach to pattem recognition is that it may continue to define extensions
of combinations of pattems that have already produced inconsistent or poor results. Allowing the computer to identify a limitless collection of pattems is just another case of overfitting, but this time at a highly sophisticated level. Heuristic programs have improved current technologj in searching, optimization, and game-plajing strategies; however, they are not readily a\ailable. There is no doubt that this technique will be quickly absorbed into trading strategies as it develops. Those readers interested in the tjpes of methods considered to be in the category of artificial intelligence, such as neural networks and genetic algorithms, should refer to Chapter 20 Advanced Techniques"). Two books of interest that represent the state of the art in heuristics and game plajing are judea Pearl, Heuristics (Addison-Wesley, Reading, MA, 1984), and MA. Bramer, Computer Game-plajing. Tbeory and Practice (Halsted Press, New York, 1983).
[start] [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25] [26] [27] [28] [29] [30] [31] [32] [33] [34] [35] [36] [37] [38] [39] [40] [41] [42] [43] [44] [45] [46] [47] [48] [49] [50] [51] [52] [53] [54] [55] [56] [57] [58] [59] [60] [61] [62] [63] [64] [65] [66] [67] [68] [69] [70] [71] [72] [73] [74] [75] [76] [77] [78] [79] [80] [81] [82] [83] [84] [85] [86] [87] [88] [89] [90] [91] [92] [93] [94] [95] [96] [97] [98] [99] [100] [101] [102] [103] [104] [105] [106] [107] [108] [109] [110] [111] [112] [113] [114] [115] [116] [117] [118] [119] [ 120 ] [121] [122] [123] [124] [125] [126] [127] [128] [129] [130] [131] [132] [133] [134] [135] [136] [137] [138] [139] [140] [141] [142] [143] [144] [145] [146] [147] [148] [149] [150] [151] [152] [153] [154] [155] [156] [157] [158] [159] [160] [161] [162] [163] [164] [165] [166] [167] [168] [169] [170] [171] [172] [173] [174] [175] [176] [177] [178] [179] [180] [181] [182] [183] [184] [185] [186] [187] [188] [189] [190] [191] [192] [193] [194] [195] [196] [197] [198] [199] [200] [201] [202] [203] [204] [205]
|