In a previous post (found here), I highlighted research discussing how low share volume on the market does not indicate any information as it relates to future returns. I also said in that post that I would expand on this research. Well, I have expanded this research and began to model price/volume relationships to objectively judge if there is any predictability of future investment returns using price/volume. What I found is that just saying volume and future price have no relationship could be a naive statement, and there may be more going on here than meets the eye.
Before I get into the results, let me preface my comments by saying when I started getting into the numbers I quickly discovered that trying to model price/volume characteristics is highly complex. In addition, potential modeling problems also arose from the open-ended nature of any time frame analysis- i.e. price/volume benchmarks may occur within the week, month, quarter, year, or more. These complexities makes simple modeling- which I prefer because I have found that complex modeling adds unknown relationships that can blow up in your face- exceedingly difficult.
With that said, I constructed a simple model that looks at volume and price relationships using criteria, that if met, a measure of future returns is calculated. In this study, I used daily prices on the SPY starting in January 1993. For any daily price point since 1993, I compared the daily price versus the previous 1 week, 1 month, and 3 month high and low price. In addition, I compared the daily volume versus benchmark volumes that occurred at the previously established maximum or minimum price levels. From this data, I created two buckets. The first bucket contained what I deemed potentially positive price/volume characteristics and second bucket negative characteristics. Positive characteristics occur when the current price trades above the previous high on higher relative volume or below the previous low on lower relative volume. Negative occurrences are logged when the price is above the previous high on lower benchmark volume or below the previous low on higher benchmark volume. The weekly, monthly, and quarterly results are presented below.
The tables above present three columns that include statistics for the buy-and-hold case, positive bucket results, and negative bucket results. The statistics I show include the the average, median, maximum, and minimum return for the various time periods. These statistics also include the standard deviation of the returns, the percentage of the returns that are greater than 0, and the Sharpe ratio (i.e. the average divided by standard deviation).
In my opinion, the 1 week and 1 month return models show encouraging results. The average and median returns for these two time period not only show a positive spread but also the buy-and-hold case lies in between both. The same is also true for the Sharpe ratios for these time periods. To me, this is evidence that price and volume relationships are indicative of future returns. That said, the 3 month case shows weaker results. Although the positive bucket exhibits a positive spread versus the buy-and-hold case, so does the negative bucket. I think this suggests that the price and volume information is more of a short-term phenomena and may not be informative for making long-term investments.
To me, the objective data says that you cannot just ignore price/volume characteristics. There is definitely something going here. That said, I would not create an entire investment strategy around these results, as the 3 month model exhibits less robust results. Although more research needs to be completed on my part, I think the data does suggest that one can use price/volume characteristics to judge entry points into certain investments.
Before I get into the results, let me preface my comments by saying when I started getting into the numbers I quickly discovered that trying to model price/volume characteristics is highly complex. In addition, potential modeling problems also arose from the open-ended nature of any time frame analysis- i.e. price/volume benchmarks may occur within the week, month, quarter, year, or more. These complexities makes simple modeling- which I prefer because I have found that complex modeling adds unknown relationships that can blow up in your face- exceedingly difficult.
With that said, I constructed a simple model that looks at volume and price relationships using criteria, that if met, a measure of future returns is calculated. In this study, I used daily prices on the SPY starting in January 1993. For any daily price point since 1993, I compared the daily price versus the previous 1 week, 1 month, and 3 month high and low price. In addition, I compared the daily volume versus benchmark volumes that occurred at the previously established maximum or minimum price levels. From this data, I created two buckets. The first bucket contained what I deemed potentially positive price/volume characteristics and second bucket negative characteristics. Positive characteristics occur when the current price trades above the previous high on higher relative volume or below the previous low on lower relative volume. Negative occurrences are logged when the price is above the previous high on lower benchmark volume or below the previous low on higher benchmark volume. The weekly, monthly, and quarterly results are presented below.
| 1 week | BnH | Vol+ | Vol- |
| Avg | 0.19% | 0.26% | 0.16% |
| Med | 0.32% | 0.34% | 0.24% |
| Max | 19.40% | 15.68% | 19.40% |
| Min | -19.79% | -10.60% | -19.79% |
| Stdev | 2.51% | 2.42% | 2.69% |
| %Up | 56.67% | 56.53% | 54.55% |
| Sharpe | 7.68% | 10.65% | 6.07% |
| 1Month | BnH | Vol+ | Vol- |
| Avg | 0.83% | 1.38% | 0.73% |
| Med | 1.38% | 1.82% | 1.24% |
| Max | 26.89% | 26.89% | 21.11% |
| Min | -29.41% | -16.00% | -27.77% |
| Stdev | 4.93% | 4.82% | 5.21% |
| %Up | 62.86% | 66.28% | 61.81% |
| Sharpe | 16.87% | 28.67% | 13.94% |
| 3Months | BnH | Vol+ | Vol- |
| Avg | 2.32% | 2.82% | 2.61% |
| Med | 3.10% | 3.87% | 3.78% |
| Max | 39.59% | 39.59% | 37.84% |
| Min | -40.51% | -26.90% | -25.17% |
| Stdev | 8.20% | 7.47% | 7.86% |
| %Up | 66.82% | 71.68% | 69.13% |
| Sharpe | 28.31% | 37.69% | 33.18% |
The tables above present three columns that include statistics for the buy-and-hold case, positive bucket results, and negative bucket results. The statistics I show include the the average, median, maximum, and minimum return for the various time periods. These statistics also include the standard deviation of the returns, the percentage of the returns that are greater than 0, and the Sharpe ratio (i.e. the average divided by standard deviation).
In my opinion, the 1 week and 1 month return models show encouraging results. The average and median returns for these two time period not only show a positive spread but also the buy-and-hold case lies in between both. The same is also true for the Sharpe ratios for these time periods. To me, this is evidence that price and volume relationships are indicative of future returns. That said, the 3 month case shows weaker results. Although the positive bucket exhibits a positive spread versus the buy-and-hold case, so does the negative bucket. I think this suggests that the price and volume information is more of a short-term phenomena and may not be informative for making long-term investments.
To me, the objective data says that you cannot just ignore price/volume characteristics. There is definitely something going here. That said, I would not create an entire investment strategy around these results, as the 3 month model exhibits less robust results. Although more research needs to be completed on my part, I think the data does suggest that one can use price/volume characteristics to judge entry points into certain investments.
No comments:
Post a Comment