检验TA信号有效性 vs. 分析TA信号内在运作机理

还在发呆

原帖由 coolboy 于 2008-4-21 14:46 发表 Very interesting discussions and it seems to me that the above inferences on how and how good TA works are convincing and probably correct. While thinking about how one can validate these inferences I ...

 

Certainly you have brought an insightful and provoking point on the table, though I would like to make some minor supplementing comments. I think a more appropriate word than “large” is “diversity” or “multiplicity”. As long as the market is not uniformed, it is less likely that someone is able to have enough resource to manipulate the price to deploy traps. Being large may imply diversity with high probability, but not necessarily. Lack of diversity is the very reason that green hands are cautioned away from small cap stock with little trading volume. It might also be valid to characterize a diversified market with slow price movement and evident trend line, because there is no manipulation and most noises are canceled away.

 

That said, it also provoked me to think more about diversity. I think most will agree that a tradable trend starts when some big player accumulates chips, and ends when the player distribute chips. The accumulation process will remove diversity from the market and distribution will restore it. Here comes the dilemma that puzzled me, because, as discussed before, lack of diversity grant the big player power to manipulate the market, and false TA signals will be emitted. So, the pessimistic conclusion I reaches is that either the trend is not tradable, or people will face false TA signals. Huh?!

 

还在发呆

#17, #18, #20, very good point， 因为简单，所以大家都用，因为大家都用，所以可靠。大道至简。

 

大家继续...

还在发呆

提两个概念，也是我刚刚想到的。

 

TA信号可以分为两种，自我满足信号和待检验信号。

 

自我满足信号指那些到了一定数值就必然会掉头的信号，因为内在机理自己就会有向相反方向运动的要求，所以叫自我满足信号。例如RSI，到了70就是超买，一定会修正。RSI(2)更有效到了98几乎会立刻回头。VIX极端高和极端低的时候，也是要回头的，因此也算是自我满足信号。胡立阳的重力线也是自满足信号。

 

待检验信号是指那些需要后续价格移动来检查的信号。比如具有心理意义的关键点位 --- 阻力/支撑等等。后续移动突破阻力位是bulish，没有突破就是bearish。但是突破之前，一般没有办法知道是否会突破。对于待检验信号，比较稳妥地操作时等检验完成之后再进场。班长一直遵守这个纪律。在未检验之前操作属于比较激进的做法，一定要做好止损。

[ 本帖最后由 还在发呆 于 2008-4-21 16:57 编辑 ]

coolboy

Did you name these new terms of “self-satisfying signals” and “to-be-tested signals” all by yourself? You sounds very creative. 

Based on what you described above, it seems to me that “range-bounded signals” and “unbounded signals” may also fit the descriptions.

So far, you have only considered “FA派/TA派二元市场模型” in the discussion. This makes us easy to understand the case of “unbounded signals” because the movement of the market in this case could be primarily determined by the two factors described above. In the case of “range-bounded signals” the so-called internal mechanism (内在机理) becomes more important than the two factors described in the above simple model when the price is getting to certain ranges.

Another example of the “range-bounded signals” is the interest rate (or the face value) of the bound market. It cannot go too high or too low and it cannot be easily manipulated by big fund managers either.

I hope my understanding is correct.

jsl

我来慢慢顶这个贴

 

Why the market has it patterns? Forget about the underlying details for now. only think about a multi-body dynamic system. The categories of participants are not so many to exhibit only statistical averages, nor so few to be easily predictable.

coolboy

A multi-body dynamical system has only a few degrees of freedom. This makes it comparison with the market, which certainly has many, many degrees of freedom, very difficult in terms of either market understanding or prediction.

还在发呆

#24

 

  thx for your nice words. actually I mean "self-fulfilling" ("自我实现") instead of "self-satisfying". I must have translated it wrongly and the wrong translation sounds creative. Certainly "range bounded" and "unbounded" are better to describe their external attributes. I wanted to focus on the internal attributes, "self-fulfilling" is active, while "to-be-tested" is passive.

 

Let me try to cook a model quickly to capture the internal mechanism of self-fulfilling (range bounded) signals.

 

还是考虑二元模型，不过现在是捕食者/猎物模型。假设最开始捕食者和猎物数量平衡。在平衡状态下，捕食者的最大利益就是尽可能的多捕杀猎物，因为捕杀的猎物越多，捕食者的相对力量就越壮大，从而就可以捕杀更多的猎物。非洲草原上的狮子和野牛就是最好的例子。少数狮子面对成群的野牛时也会束手无策。所以在平衡状态下，捕杀的越多越有利。

 

可是当捕食者的相对数量多到一定程度的时候，继续捕杀猎物就变得很困难了。因为缺乏食物，老弱病残捕食者就会被淘汰。而幸存下来的猎物多是年轻力壮的，他们繁衍出的后代也会继承这些优良基因，使得猎物的队伍开始壮大。自然而然的，天平开始逆转，又回到平衡状态。

 

比较FA/TA模型和捕食者/猎物模型，我觉得前者描述市场的投资性。FA派知道股票的真实价值，TA派虽然不知道，但是通过模仿FA派，也能达到股票真实价值。后者描述市场的投机性。在大部分时间，你应该站在捕食者一边，在极端情况下，你应该站在猎物一边。

 

#25 and #26

 

I think multi-body system is a term in physics. in social science, it's adaptive system or complex system --- a system with vast number of paticipators each of whom has independent thinking and decision-making.

 

the real problem is that as the number of paticipators increase, the complexity of the problem increase exponentially. Predicting the next step of such system is intrinsically hard, if not impossible.

 

however, there are still chances for us to predict, when certain component in the system rised to dominate other parts, which will significantly reduce the complexity. for example, financial crisis certainly dominate other parts in the stock market recently.

 

越来越有意思了，大家继续。 

[ 本帖最后由 还在发呆 于 2008-4-22 03:15 编辑 ]

武当人

再来掺和一下，顶一顶楼主的好帖子。

楼主用二元模型来解释市场而jsl用多体模型来解释市场。结果coolboy说楼主是“very creative”而说jsl是“very difficult”。难道“多”不如“二”好？


[ 本帖最后由 武当人 于 2008-4-22 08:03 编辑 ]

jsl

我的意思是说 理解市场为什么有pattern是重要的， 先不必去追究机制的细节。 当然要把握机制的正确方向。 学EE的可以考虑一个略复杂的信号发聩体系。学统计的可以考虑一个Baysian network. 学计算机算法的可以考虑planning algrorithm or Markov decision process. 学物理的可以用mean field theory. If you are a specialist in certain field, you might think of something like fricition dynamics etc.

 

二元体系是重要的，应该能解释局部现象。Good to be used for local approximation. 事实上很多问题的解决基于二元体系。 Divide and conquer

 

Based on the obseration of market pattern, I can claim that the effective degree of freedom is not high. There are many variables, there are also many contraints. Unconstrained system with high degree of freedom only exhibite average statistical behavior.

 

书院是学物理的，给大家讲讲。

CoolMax

还有一个非常棒的指标：

量加权的平均价格。和MA不同的地方是价格用交易量加权，做DT很有用。观察分钟线，你会看到，股价在这条线上下，受到支撑或阻力。成为很好的进入点。 AlphaTrends的Brian常用。

可惜StockCharts和TD都没有这个指标。

还在发呆

#28

 

模型有各种用途，有用来分析问题的，也有用来计算，模拟的。简单的模型适于分析问题。复杂的模型不光分析起来费力，计算起来也很复杂。但是如果要想得到接近实际情况的计算结果，模型势必要复杂到接近实际情况。用二元捕食者/猎物模型也可以进行计算，不过估计算出来的就是一个阻尼正弦波，和市场走向相比可就差的远了。我觉得两者没有绝对的优劣，看具体的应用而定。

 

#30

 

赞！volume-weighted average price能更好的反映平均成本。

 

#29

 

把这个留到最后，因为我想把他当成下面讨论的重点。先说点不同看法

Based on the obseration of market pattern, I can claim that the effective degree of freedom is not high

我觉得有必要区分number of patterns和degree of freedom of system。pattern个数少，不一定说明系统自由度低。考虑大气模型，每个分子都可以在3维空间和一位时间上运动。自由度是分子总数的4次方。可是大气的宏观pattern却很简单，基本上都是环流，大到季风，中到飓风/台风，小到龙卷风。

 

或者jsl的意思是需要构建一个以pattern为核心的模型？这个模型只考虑pattern和pattern之间的转换？这样的话自由度就小很多了。

 

愿闻其祥。

 

jsl

其实就是macro economics and micro econimics 的关系。At aggregation level, there arn't too many driving forces. In a relatively short period, changes are more important. That's why I say the “effective" degree of freedom is not high. 也因此市场才会出现一些有序现象。这个问题有些philosophical, 理解就好了，大家的想法没有什么本质区别。

 

 In markov decision process, there are two things you need to pay attention to, the current state and transition matrix. Do you think you have constantly ignored one?

coolboy

It is interesting to think about the stock market as a Markov process. It is also true that if we have long enough data series we will be able to calculate the transition matrix. This may well correspond to the utility of “pattens” in TA analysis. One small catch about this comparison is that one has to assume that the market is a stationary process, i.e., it only fluctuates close to a constant or a slowly varying mean state. Therefore, TA that is essentially based on Markov process will most likely work well if the market does not turn rapidly.

st-watcher

What is the English name for 量加权的平均价格? i can try my Trade Station and ThinkorSwim's Prophet Chart to see if they have it.

Thank you

CoolMax

Institutional Trading Benchmark

Volume Weighted Average Price (VWAP)

People often wonder why the trading volume slows so dramatically during the middle of each trading day. The most common explanation for the midday inactivity is that it is lunch time and while it is certainly part of the reason, it is not as though the specialists and market makers all head out to a liquid lunch and forget about their business for the day. Most serious traders I know eat their lunch at their desk so they don’t miss an opportunity and to maintain a better feel for how their positions are trading.

Consider a market maker with a day order to buy 1 million shares of a stock for an institutional customer. The market maker cannot just go in and buy the full position in the first two hours of the day and then leave his desk to go play golf. Market makers are evaluated by their customers for the quality of trade execution; did they get a fair price for the customer? The most common method of used to analyze the quality of a trade execution is to compare the price the order was filled to the Volume Weighted Average Price (VWAP). The VWAP is calculated by dividing the dollar volume of a stock by the share volume over a given period of time. Simply put, the VWAP is the average price that each share was executed over the period of time being studied. There are several ways to calculate the VWAP in RealTick, the analysis and trading software I use. I prefer VWAP analysis with a moving average, in particular for shorter term day trades on a one day chart of the equity I am trading.

The VWAP is considered to be a fair benchmark for comparison of the institutional trade desks ability to execute trades on behalf of the customer. If the brokerage purchases were made at a price less than VWAP then they did a good job for the customer. If the price paid was greater than the VWAP they may lose that customer for future transactional business. The daily VWAP is a number which changes as orders are transacted at varying prices throughout the day.

Here is a simple example of how an institutional trader might manually execute an order for the purchase of 1 million shares. Let’s say the stock closed the previous day at $40/share. The morning the broker receives the buy order he may offer 5 or 10,000 shares at 39.90 or lower while simultaneously bidding for shares at a lower price. The first trick this broker may use is to show the full size of the offer while only showing 100 shares bid with a larger number of shares “in reserve” meaning that he may be really bidding for 5000 (or any other number they choose) shares. By showing a larger number of shares for sale and a small amount of demand in the pre-market, the broker may induce weaker holders to sell their shares in fear that there could be a real seller looking to get a head start on their
selling that day. This type of activity is pure manipulation and it happens all the time!

[ 本帖最后由 CoolMax 于 2008-4-22 21:32 编辑 ]

jsl

VWAP is an important reference for institutional trading, so it is important

usami

原帖由 CoolMax 于 2008-4-22 14:01 发表 还有一个非常棒的指标：量加权的平均价格。和MA不同的地方是价格用交易量加权，做DT很有用。观察分钟线，你会看到，股价在这条线上下，受到支撑或阻力。成为很好的进入点。 AlphaTrends的Brian常用。可惜StockCh ...

 

 

叫啥名字？

CoolMax

原帖由 usami 于 2008-4-22 21:34 发表     叫啥名字？

 

VWAP,

楼上有解释。

还在发呆

#32 & #33

 

It took me a while to fully digest the similarity between Markov transition matrix and patterns in stock market. For most of the state, the corresponding column in the transition matrix must be very complex, thus it is able to generate the vast possibility we observed in stock market. However, for certain state, I shall name it “critical state”, the corresponding column is very simple, containing only a couple of non-zero entries. At these critical states, it is much easier to predict the future movement. Patterns are just the critical states that mandate a few number of certain future movements. Being a trader, one needs to be patient to trade only those critical states.

[ 本帖最后由 还在发呆 于 2008-4-23 03:02 编辑 ]

jsl

We can not assume the Markov process is stationary, that's why we need some FA work and keep updating the estimate on transition matrix estimations ( we don't have to come up with something accurate). Also we should never assume that we know the current state, this goes to the area of POMDP (google it) and you must use a blieve state. At least, "critical states" IS what you want to find out!!!