In the 2011 academic paper “Another look at trading costs and short-term reversal profits“, the authors (De Groot et al) speculate that a simple mean reversion strategy in US stocks is able to significantly outperform the market.

The strategy seems to have promise, especially when used in a rotational setting with smart allocation.

Trading costs and short-term reversal profits

According to the authors, there is a growing body of literature that supports something known as the short-term reversal anomaly. In other words, the idea that stocks with relatively low returns over the past month or week, earn abnormal returns in the following month or week (and vice versa).

However, most of the academic literature on the subject finds that those returns are cancelled out once transaction costs are taken into account:

“Conrad, Gultekin and Kaul (1997) report that most of short-term reversal profits fall within bid-ask bounds and Keim and Madhavan (1997) find that reversal strategies require frequent trading in disproportionately high-cost securities such that trading costs prevent profitable strategy execution.”

The authors of this paper suggest that the impact of trading costs on profitability is largely attributable to trading in small cap stocks. In other words, short-term reversal strategies CAN be profitable, so long as they are limited only to the largest, most liquid companies.

By limiting the stock universe to large cap stocks and by applying a more sophisticated portfolio algorithm, a simple reversal strategy is shown to generate 30 – 50 basis points per week net of costs.

“Several studies report that abnormal returns associated with short-term reversal
investment strategies diminish once trading costs are taken into account. We show
that the impact of trading costs on the strategies’ profitability can largely be attributed
to excessively trading in small cap stocks. Limiting the stock universe to large cap
stocks significantly reduces trading costs. Applying a more sophisticated portfolio
construction algorithm to lower turnover reduces trading costs even further. Our
finding that reversal strategies generate 30 to 50 basis points per week net of trading
costs poses a serious challenge to standard rational asset pricing models. Our findings
also have important implications for the understanding and practical implementation
of reversal strategies.”

Basic short-term reversal strategy & methodology

The basic idea behind this strategy is to hold a portfolio of both the strongest and weakest stocks for a short period of time. Stocks are ranked by liquidity and performance and placed into relevant deciles so that the largest, strongest stocks are bought, and the largest, weakest stocks are sold.

The authors provide evidence to suggest that this strategy is unprofitable when trading small cap stocks and European stocks, because of the higher transaction costs involved. Furthermore, the strategy performs poorly on the top 1500 and top 500 US stock universes thanks to the wider spreads involved with these securities.

However, the strategy is shown to work well on a universe of the 100 largest stocks and shown to produce profitable returns of 30 – 50 basis points a week.

Transaction costs

In order to test this strategy, the authors collect data on transaction costs from two different sources; a volume based model dating back to 1997 from Keim and Madhavan, and a model provided by Nomura Securities.

Shown below you can see the various transaction costs as suggested by the two models. So, you can see that average transaction costs for the top decile of the largest 1500 US stocks between 1990 and 2009 was 8 basis points, or 0.08%:

nomura transaction costs

Using these estimates and applying the strategy as outlined in the paper, the authors present the following results. You can see that for the 100 largest US stocks there is a total return of 31.5 basis points for the long-short portfolio and 17 basis points for the long-only portfolio:

short term reversal strategy results

Testing the strategy in Amibroker

At first glance, the returns shown in the table for such a simple strategy seem too good to be true. If markets are nearly efficient, and we are trading in the most efficient stocks, it seems unlikely that we would be able to produce such abnormal returns.

I therefore opened up the trading simulator from Amibroker and attempted to run some tests of my own.

Test one

For this first test, I loaded up the S&P 100 universe from Premium Data (which includes historical constituents) and instructed Amibroker to go long the weakest 10 stocks over the last five days. Stocks were weighted equally, with a starting capital of $100,000 and the portfolio was rebalanced every 5 days.

Similar to the paper discussed, the strategy buys the weakest 10% of stocks in the universe, holds those stocks for five days, then replaces them with the weakest 10%.

Running this strategy on the S&P 100 universe between 1/1/2010 and 1/1/2015 without any transaction costs produces a compounded annual return of 10.18% with a maximum drawdown of -15.89%:

CAR: 10.18%
MDD: -15.89%
CAR/MDD: 0.64

short-term reversal strategy equity curve

Introducing transaction costs

In the paper, transaction cost data is provided from Nomura Securities and these figures suggest an average transaction cost of 6-8 basis points for the top 100 US stocks. I round this figure up to 8 (or 0.08%) and I re-run the test.

Using transaction costs of 0.08% per trade (plus $1) produces a compounded annual return of 5.12% with a max drawdown of -18.63%.

CAR: 5.12%
MDD: -18.63%
CAR/MDD: 0.27

short-term reversal strategy equity curve plus transaction costs

How do these numbers compare

The authors suggest a long-only strategy similar to this provides a return of around 10% annually but our strategy indicates that these results may be optimistic. Moreover, the compounded annual return of 5.12% between 2010 and 2015 is significantly lower than the benchmark buy-and-hold return during the period, which is 11.89% with a max drawdown of -17.67%.

Introducing the smart allocation technique

The authors of the paper are able to improve returns significantly by introducing what’s called a smart allocation strategy.

Instead of rebalancing the portfolio every five days, stocks are held on to, so long as they are still in the top 50% poorest performers. As soon as a stock drops out of the top 50% of poorest performers, it is sold.

This means positions are held slightly longer and the turnover is reduced (thus reducing the impact of transaction costs). We can do a similar thing in Amibroker by utilising a rotational strategy.

Test two

In test two, we use the rotation function in Amibroker and instruct Amibroker to buy the 10 poorest performers in the S&P 100 universe. Each position is held until it drops out of the worst 50 performers.

For example: on the 12/17/2014 $AVP is the second worse performing stock in the universe having fallen almost 8% over the last five days. The stock is bought and held. On 12/31/2014, the stock has rallied and is no longer in the bottom 50 market performers. The stock is then sold for a 3.44% profit and a new position takes it’s place.

AVP example long trade

Running this test on the S&P 100 universe between 1/1/2010 and 1/1/2015 (and keeping transaction costs at 0.08% per trade) produces an annual return of 16.63% with a maximum drawdown of -29.57%.

CAR: 16.63%
MDD: -29.57%
CAR/MDD: 0.56

short term reversal strategy equity curve

Next I move the strategy back in time to see how it coped during the credit crisis and over a longer time period. Running the strategy between 1/1/2000 and 1/1/2015 produces an annual return of 18.28% with a maximum drawdown of -55.78%.

CAR: 18.28%
MDD: -55.78%
CAR/MDD: 0.33

short term reversal strategy equity curve

profit table for the short term reversal strategy

short term reversal distribution of results


We were not able to replicate the short-term reversal strategy from the academic paper completely and our results do not appear to be quite as strong as reported in the paper.

However, when using the smart allocation strategy the results do show promise. They are dependent on tight transaction costs so traders should prefer not to use this strategy during times of fast markets. (During these times, tight spreads can not be guaranteed).

When transaction costs are increased to 0.2% per trade CAR drops to 9.37% and maximum drawdown increases to -61%. A volatility filter could therefore work well to limit trades during choppy waters.

Tests performed with Amibroker using data from Norgate Premium Data.

For more trading strategies and systems click here.

want the code button image

If you would like the Amibroker AFL code for this strategy just pop your email in the box below and you’ll be able to download it on the next page.

Past performance is no guide to future returns, accuracy is not guaranteed. Please read the full disclaimer.

Tags: , ,

5 opinions

    • Ola

    • July 9, 2015

    • 2:15 pm

    • Reply

    Very interesting post!
    May I ask how you are managing the survivorship bias in the back tests? Would that be a potential cause for the discrepancies between the results?

    Best regards,

    • Hi Ola
      I used the S&P 100 universe from Premium Data which includes historical constituents. So there shouldn’t be any issue with survivorship bias.
      That said, I didn’t do any stress-testing on this system so there may well be some discrepancies somewhere else.
      Thanks for your message.

    • Ying

    • December 4, 2015

    • 7:55 am

    • Reply

    Hi Joe,

    I’m curious which journals do you keep an eye on? I’m new to this area and I’d like to see some new researches on trading strategies, could you please recommend some? Thanks!

    • Hi Ying,
      I like to keep an eye on Quantpedia is good but it costs money.
      Actually, I am planning to write an article next week of places to find quant strategies so keep an eye out!

        • Ying

        • December 5, 2015

        • 11:42 am

        • Reply

        Awesome, looking forward to that!

Leave a Reply

%d bloggers like this: