Efficiency of Indian Equity Markets
This report examines the efficiency of Indian Equity Markets when they were in a bull run. The statistical results are inconclusive but the forecasting accuracy of some of the techniques is incredible.
Some of the conclusions that can be drawn from this work are:
- News is the key driving factor in any stock market. The inherent randomness of the prices comes from the unpredictable flow of information. Market participants may react differently to the same information and may struggle for a while before an equilibrium is reached. This partly explains the high volatility in the opening and closing hours.
- As far as econometric analysis goes, there is absolutely no need to consider the high, low, open and close series separately. They have the same information content. Picking up one of these and consistently will yield the same results.
- Important stock indices move in unison. That is there is hardly any arbitrage oppurtunities across indices. This implicitly implies efficiency of the markets. The important indices are cointegrated. This means that there is same underlying data generating process affecting all the markets simultaneously. This data generating process has to be flow of information.
- In spite of the fact that indices are cointegrated and fast to react, the investors may benefit from the continuity of the series. This may be compared to inertia associated with mass.
- Stock prices are “continuous” series in the sense that prices hover around the past prices except under extremely unusual conditions. This clubbed with the efficient market hypothesis gives a lot of
credibility to random walk model. - Random Walk can never be used to make any profits at any point in time. This is easily the poorest forecasting technique. In the last 10 years and across five major indices, there was never ever a single instance when this model forecasted correctly.
- The Q statistic at various lags confirms that the null of no serial correlation can be rejected for all the indices. This is not in accordance with efficient market hypothesis and is a proof to reject the weak form of efficiency hypothesis.
- The unit root tests support the random walk model. But instead of saddening, this should make the speculators happy because it means that the near about range of future prices are known.
- The risk assosciated with continuing in the market increases linearly with time. It is five times more risky to remain invested for five days than it is for one day.
- The non-trading days are not risk-free but there risk is not same as remaining invested on a trading day. This observation directly follows from the skewness in variance ratios at lags more than five.
- The market does have a long term memory. The Hurst Exponent examination tells us that the prices upto more than one lag have significant impact on tomorrow’s price.
- The investor must note that on any day a market is as likely to close high as it is to close low. This is a direct result of the runs test. This does not in anyway attack the predictability of the direction, it simply confirms that the market continuously corrects itself and comes to equilibrium.
- The above observation is the motivation behind trend analysis. Trend analysis was found to be fruitful only for the price series not for the returns because the returns for the most part are heavily clustered around a mean.
- Trend analysis was found to be a superior forecasting technique than random walk model. Whether a market is a random walk or not may interest the academics. For the ordinary investor, trend analysis is much better than random walk.
- Quadratic trend analysis is superior to both linear and exponential trending techniques. This suggests a lot of non-linearity in the market prices and is a strong motivation to look for non-linear forecasting models.
- Artficial Neural Networks were used to capture the non-linearity in market variables. ANNs proved to be a better forecasting tool than both trend analysis and random walk.
- ANNs were found to be quite sensitive to the number of outputs forecast. It is better to predict each output separately. When I tried to simulatenously predict the direction of movement and value of closing pricing, the MAPE and MAD deteriorated as compared to the cases where these variables were forecast separately.
- ANNs were able to forecast the shape of the price series quite accurately for upto 70 trading days in future. There was a scaling problem and hence there is a scope to improve upon the MAPE, MAD and MSE numbers.
- ANNs are extremely sensitive to nature of scaling function. For optimum results there should be scope for the output to go beyond the current maximum value. This has been a shortcoming of this study.
- In sum, the investors can do better than random walk by using these forecasting tools. The outputs from these tools with the subjective and qualtitative assesments (read forecasts), should land the speculator in much better situation than just relying on random walk.
The report can be downloaded from here. Slides of the final presentation are also available.
Of course all your investment and speculative decisions are YOURS. I bear no legal or moral responsibility for your actions.
Hi,
This is an excellent piece of work! Good job!! Can you please also share the link of the zip file of all the supporting docs, as you mentioned in the end notes section of your pdf file. I am greatly interested in carrying your work forward.
Regards,
Pankaj.
Posted by Pankaj on November 18, 2008 at 12:09 AM TPT #
Hi,
This is an excellent piece of work! Good Job!! Can you please also provide the link of the zip file of all the supporting docs as mentioned by you in the end section of your pdf file. I am greatly interested in carrying further research on the subject.
Thanks and regards,
pankaj
Posted by Pankaj on November 18, 2008 at 12:14 AM TPT #