Hi John!

i don't see any obstacles for transition from Bull to Bear market strategies intra hour.

First trick; define bull market and define bear market without resorting to the nonsensical and arbitrary percentage movement from tops and bottoms often cited. There is up and there is down. There are upward tracking trends and downward tracking trends. Algos are agnostic. Math and statistical systems are agnostic. They can distinguish between up and down. What more does anyone need? Finally, there are no predictive models-none.

Algorithmic trading model should have an in-built strategy for the bear market - meaning that the model should identify the bear phase automatically. The real challenge though is the ability to detect sooner than later, assuming you want to follow the trend. There could be a model which may go against the trend as well. However detecting the bull and the bear phase is imperative in either case.

Once again without appearing obtuse, can the terms be defined. What is a bull and what is a bear? Also, how can a model be built against trend? Is it valuation based, time cycle, sunspots, etc.? If the trend is rising, detecting reversals is a subjective artistic affair rather than mathematical. Merely curious and trying to learn.

I used the term bull and bear to mean what it normally means. When markets move upwards between point A and point B the trend is bullish and vice versa. So during bear phase if the model suggests going long, then it is going against the trend.

Please see the article posted in linkedin https://www.linkedin.com/pulse/consistent-performance-from-stock-markets-ramachandra and https://www.linkedin.com/pulse/algorithmic-model-promarvel-part-2-venkata-subramani-ramachandra?trk=prof-post where I have provided the details of the results back tested for about 13 years. The model is a buy-hold-sell strategy with no shorts. 2008 is obviously a bearish year and the markets started looking up only from mid 2009. Clearly this is a model that works against the trend.

Nothing is obvious in the capital markets. Regarding movement from Point A to Point B, what is the time period and what is the price dimension? When and what marks the end of a cycle? Unfortunately the depictions are always, emphasis on always, designated via retrospective observation. The system in place here applies trend data hence I understand the basis for the model work product cited above. The perpetual challenge is preventing paralysis through analysis. One of the cruel hoaxes perpetrated by the stock markets averages is that they tell us very little or nothing until as forward indicators. Turns are commonly long underway prior to the averages providing meaningful clues regarding the flow of events. Substantial capital is often destroyed awaiting a signal from the averages. The leaders attracting the most love and attention roll south but the common view is that the averages remain relatively intact so the declining sectors are merely undergoing a "correction" prior to advancing once again. Correction is one of the more absurd terms thrown around the Street. Price is always correct. Detecting major trend shifting is largely an art that may be supported by data. The data allows us a clearer visualization of prevailing conditions. The fifth major bear market of my tenure in this business is already developing. As always, the timing is impossible to forecast. The previous four possessed characteristics unique to the episode with causation debated to this day. Many underlying conditions approaching the onset were shared buy many were not. The same applied to upside reversals ad bear market terminations. Perverse circumstances occurred more often than not. For example, the bear market developing on the backside of the growth stock nifty fifty mania of the late 1960' and early 1970's obscured opportunities developing underneath the glare of the distressed popular averages. The situations largely constituted the leadership of the subsequent market recovery years later. As a sell side broker during the period we needed to dig around to scrounge out a living and stay in the business. The popular averages were always a concern but could not be allowed to rule our lives. On balance the outcomes were constructive for salespeople and customers alike. We were tasked by clients to find situations that represented sensible business propositions that could work. The market averages were interesting but no excuse for failing to deliver. Returning to the discussion subject directly, the art component of market work was trivialized considerably in recent years. I have a suspicion that it will regain some acceptance and respect as most of the quantitative models fail us all once again as they have in the past. There is a sense that CA Venkata infused considerable intuitive elements based upon professional and personal experience into the models constructed. As such, The detection of counter trend prospects and activity foreshadowing prospective major changes persuades me that he is doing something more than plotting lines and running formulae absent any subjective elements embedded into the model.

Thanks Larry for your detailed comments. I really admire your depth of knowledge and your insight into the markets having been a witness to the market cycles several times - at least four of them being major. My firm opinion is that there is constant mispricing in the equity markets. My research and consequent development of this model [which is spread over a period of the last 11 years] is applicable only for equity stocks - and that too the ones that qualify to be part of the top Index - in India for example the Nifty Fifty stocks. The mispricing is on account of several factors including the ones that are artificially created by market manipulators with or without inside information. The objective of this model is to exploit the mispricing to our advantage. I am not sure if this approach has been considered by anyone else in the way that I have approached the issue. I have developed around 20 parameters that help identify a) the stock b) the quantity in relation of the size of the portfolio [which according to me is the most important factor] c) when to buy additional quantity or sell part or full quantity. The quantity is derived based on the co-efficient of four proprietary surface charts that I have developed. The essence of these multipliers is that more allocation is made in those stocks that is expected to exit at a predefined profit soon. The IRR on each such investment is well above 150% while the ROI of the portfolio hovers around 22% p.a. on the minimum while back testing for the last 13 years. There are no subjective elements built in the model. I will be writing more about this model in Linkedin Pulse posts regularly in the days to come.

Algos allow trading without concern for bull or bear markets. You may use a measure (any measure you may think of) to bias the algo in a long or short trading direction in order to improve results. You do not need to trade short in bearish markets or long in bullish markets, you can go both ways. It all depends on how you frame your trading system. An algo will not provide answers for your trading system, it works the other way around, you must provide the answers for your algo.

CA Venkata, thanks for the compliment. Simply sharing some experiences and random ramblings. Your allocation work strikes me as attending a highly underdeveloped aspect of the business; genuine risk adjusted allocation. A version is used here to manage size and balance but does not combine elements in the manner yours obviously does. My front end selection process also measures factors such as the company's return of capital and earnings productivity relative to itself rather than relative to the general market. There are periods, such as the current time frame in the USA, when the number of qualifying prospects is relatively small. Essentially, the market takes us out of a market. The issues that do qualify are weighted according to price risk present upon commitment relative to the mandatory sell stop which is a dynamic figure. Position management is performed according to the price risk factor as well as continuing compliance with the selection process. Pursue your project intensively. It has a place in this world. Oscar's remarks are a reminder that bull or bear is less relevant than thought. The objective is identifying prospects that can work and first assessing them independently of the market condition. Since no one knows what happens next, the constant search for prudent situations is a must. If wrong, the numbers will tell us soon enough and take it away. If no effort is made then there is no opportunity to gain.

You can experiment with what are called Regime Change Markov Models. If you can define a regime, the model can tell what was the previous regime and what is the current regime. It can also give you transition probability estimates (absolute or conditional) from one regime to another.

You may want to see my next post in the Pulse on Algorithmic trading. Please do let me know your views:

https://www.linkedin.com/pulse/algorithmic-model-promarvel-part-3-ca-venkata-subramani-ramachandra?trk=prof-post

Thanks.

A lot depends on the time scale that you are looking at. From what I have seen in energy futures, not all participants react at the same time and the same rate. There is an "information" inequality and an "act on the information" inequality that can be detected in the presence of arbitrage cycles, i.e. programmatically by an algo trading system.

The relationship between arb cycle properties and price/volume movement might tell you something, but at this point I couldn't tell you what.

Would you like to detect it after the fact or would you like to detect that it's coming?

Well, if the algorithms go long and short then it shouldn't matter. Also, if the entire portfolio management methodology contains iteration (say every 6 months) on new data then it shouldn't matter either. My experience is mostly stocks and corresponding options.

Tom,

It does matter for long short strategies too. If your long short strategy is based on the trend following philosophy, sideways markets are dangerous. It is very very useful to detect a regime, assuming its possible.

Hi Bharath, What you say is true. It's difficult to give answers to these types of questions because there's more unsaid. I'm making a remark with assumptions with your line of thinking included. What you call regime i call iteration. Then if the algo is long and short then it shouldn't really matter. Just feed new information to let the algo adjust.

A long short strategy which is based on some kind of proven mean reversion hypotheses, can still work in sideways markets, but strategies based on trend following will find it very hard

Agree

Can't be done. You're talking about 'hey can we time and guess where the market is going to go'. Until someone changes my mind I;m convinced market movements are completely unpredictable.

Have you heard of the Hindenburg indicator?

http://www.investopedia.com/articles/trading/07/hindenburgomen.asp

Thanks, David

I do not understand the fascination and application of mean reversion. I see numerous references in numerous places. However, no one ever describes the time slice selection process or how to run a meaningful proof application. Perhaps mean reversion can be proven but to which mean and when. Means are constantly moving. Running a scheme with a fluid price against a frozen mean is simply delusional and statistical inappropriate. As mentioned previously, give a monkey a typewriter and eventually he will write the sonnets of Shakespeare.