A Comparison of Short-Term Mean-Reversion

Indicators for Global Equities

Raymond C. Micaletti *

ABSTRACT

We examine an array of short-term mean-reversion indicators for global equities. The

indicators encompass the most widely known price oscillators from the field of technical
analysis along with several modified versions first developed by the author in the 2009-
2010 time frame. Constructing simple trading strategies from a wide range of indicator
parameters, triggering thresholds, and holding periods, we find the modified oscillators
tend to dominate the performance rankings on both the long and short sides of the market.
Consequently, this study may serve as a point of reference for day traders, swing traders,
or even asset managers looking to better time their rebalances.

* Co-founder and the chief investment officer of Relative Sentiment Technologies, LLC.

Electronic copy available at: https://ssrn.com/abstract=4339128

Short-term mean-reversion—i.e., the tendency for short-horizon market moves to reverse,
partially or fully, over subsequent short horizons—is an empirical fact of most equity
markets. But this wasn’t always the case.

In fact, up until the late 1990s, the U.S. equity market (along with most others) was
momentum-driven—up days tended to follow up days and vice versa. During the bursting
of the dotcom bubble, however, this momentum-type behavior gradually transformed into
mean-reverting behavior, where up days tended to follow down days and down, up. It’s
been that way ever since.

We can visualize this behavioral shift in two ways. The first is by charting a quantity
known as the “variance ratio” (Lo and Mackinlay (1988)). This ratio measures the variance
of N-day returns divided by N times the variance of 1-day returns:

Var (r N , M)
VR( N, M) =
N × Var (r1 , M)

where VR(·) is the variance ratio, Var (·) is the variance, r N is the N −period return and M
is the length of time over which the variance is computed.

If the time-series is random, i.e., neither mean-reverting nor mean-averting (trending), the
variance ratio will equal 1. If the variance ratio is greater than 1, the variance of N-day
returns is greater than N times the variance of 1-day returns—a hallmark of mean-aversion.
A variance ratio less than 1, conversely, signals mean-reversion.

Figures 1 and 2 illustrate this concept using a two-period binomial tree. If one moves
up-up or down-down in the tree, the two-period return is ±2R, while the one-period
return is ±1R. The variance of the two-period return is 8R2 while the variance of the
one-period return is 2R2 . This results in a variance ratio of 2, which is greater than 1 and
thus indicative of trending behavior.

In contrast, if one moves up-down or down-up, the two-period return is 0, resulting in a

variance ratio of 0. As 0 is less than 1, this signals mean-reversion.

Figure 3 shows the 200-day moving average of the total U.S. equity market’s variance ratio
from 1927 to 20221 . We use a 5-day return in the numerator and a 63-trading-day lookback
(approximately three months) to compute the variances (i.e., N = 5 and M = 63).

1
Data taken from the Kenneth R. French Data Library French (2022)

Electronic copy available at: https://ssrn.com/abstract=4339128

Figure 1. Binomial tree offering the possibility of either an up-up move or a down-down move over two
periods. This behavior signifies mean-aversion.

Figure 2. Binomial tree offering the possibility of either an up-down move or a down-up move over two
periods. This behavior signifies mean-reversion.

Electronic copy available at: https://ssrn.com/abstract=4339128

Figure 3. 200-day moving average of the variance ratio (VR(5, 63)) for the U.S. equity market
from 1927 to 2022.

From Figure 3 we see that from 1926 until roughly 2000, the variance ratio stayed mostly
above 1, an indication of a trend-friendly U.S. equity market. From 2000 onward, however,
the ratio has generally stayed below 1, signifying the existence of mean-reversion.

Another way to visualize this shift is to plot the performance of two basic trading strategies:
one that buys the market at the close if that day’s return is positive and one that buys the
market at the close if that day’s return is negative.

Figure 4 plots the (frictionless) performance of these two strategies from 1926 until March
24, 2000 (the top of the dotcom bubble). Figure 5 plots the performance from March 25,
2000 until December 31, 2022.

As is evident, the momentum strategy was ultra-dominant until 2000, after which point
the mean-reversion strategy became ascendant. The timing of this shift lines up well with
the aforeseen shift in the variance ratio.

Electronic copy available at: https://ssrn.com/abstract=4339128

Figure 4. One-day momentum and mean-reversion strategies, July 1, 1926 - March 24, 2000

Figure 5. One-day momentum and mean-reversion strategies, March 25, 2000 - December 31, 2022

Electronic copy available at: https://ssrn.com/abstract=4339128

While the existence of short-term mean-reversion in modern equity markets appears
widespread, its causes remain elusive. Two primary theories have been offered (Da et al.
(2014), mainly in the context of single-stock price reversals over one-month horizons):

1. Investor overreaction: Many researchers (Shiller (1984), Black (1986), Stiglitz (1989),
Summers and Summers (1989), Subrahmanyam (2005)) have suggested that investors
overreact to information or otherwise fall prey to cognitive errors, which causes
prices to overshoot and then subsequently retrace. (This would not appear to explain
multi-day mean-reversion at the broad-market level, however.)

2. Compensation for providing liquidity: Others have suggested (e.g., Campbell et al.
(1993)) that market makers provide liquidity at concessionary prices to uninformed
traders seeking liquidity. The subsequent reversal in price, then, serves as compensa-
tion to the liquidity-providers.

Because short-term (i.e., several-day) mean-reversion is a rather recent phenomenon, this

latter theory would suggest a rise in the number of uninformed traders since the turn
of the century. Such a rise would have coincided with the rise of day trading via online
brokerages as well as the ever-increasing amount of liquidity sought out by employee-
retirement plans and ETF rebalances.

Another potential cause might be the fund flows arising from the hedging of index options
by options dealers. Most of the time these dealers are “long gamma” (SqueezeMetrics
(2017)), which results in dealers having to sell more of the market as the market rises and
buy more as the market falls. This dynamic provides a stabilizing force to the market
in favor of mean-reversion. Perhaps the index options market didn’t provide enough
market-moving flows until the year 2000 (or perhaps prior to that time the natural state of
options dealers was to be “short gamma”)?

Whatever the causes of short-term mean-reversion, exploring these theories in detail

is beyond the scope of this paper. Instead, we focus on identifying which short-term
mean-reversion indicators work best across global equity markets.

To that end, the rest of the paper is organized as follows: Section I discusses the universe,
data sources, and indicators included in the study. Section II outlines the trading strategies
and metrics used to compare the indicators. Section III presents the results and Section IV
summarizes the findings.

Electronic copy available at: https://ssrn.com/abstract=4339128

I. Universe, Data, and Indicators
We consider the following universe of global equities (as represented by dollar-based
ETFs):

1. SPY: S&P 500

2. QQQ: Nasdaq 100
3. MDY: S&P 400 (mid cap)
4. IWM: Russell 2000 (small cap)
5. XLE: S&P 500 Energy Sector
6. XLB: S&P 500 Materials Sector
7. XLI: S&P 500 Industrials Sector
8. XLY: S&P 500 Consumer Discretionary Sector
9. XLP: S&P 500 Consumer Staples Sector
10. XLV: S&P 500 Healthcare Sector
11. XLF: S&P 500 Financials Sector
12. XLK: S&P 500 Technology Sector
13. XLU: S&P 500 Utilities Sector
14. IYR: S&P 500 Real Estate Sector
15. EFA: MSCI Developed Markets Equities
16. EEM: MSCI Emerging Markets Equities
17. EWJ: MSCI Japan

These dollar-based ETFs represent different capitalization classes (large, mid, and small),
U.S. sectors, developed and emerging markets, and Japan.

We use 1-minute open-high-low-close-volume (O-H-L-C-V) data, adjusted for splits and

dividends, to compute both the indicators and the daily returns used to measure the
indicators’ performance. We obtain this price data from Tiingo.

In what follows, we will refer to the O-H-L-C-V values at time t as Ot , Ht , Lt , Ct , and Vt ,

respectively.

We examine the following indicators:

1. Relative Strength Index (RSI): Perhaps the most widely known of all price oscil-
lators, RSI was introduced by J. Welles Wilder in 1978 (Wilder (1978)). It is given
by:

Electronic copy available at: https://ssrn.com/abstract=4339128

100
RSIt = 100 −
1 + RSt
WMA(Ut , N )
RSt =
WMA( Dt , N )
1
WMA( Xt , N ) = WMA( Xt−1 , N ) + ( X − WMA( Xt−1 , N ))
N t
Ut = Ct − Ct−1 if positive, 0 otherwise
Dt = Ct−1 − Ct if positive, 0 otherwise

WMA( Xt , N ) represents an N-period Wilder Moving Average (initialized with an

N-period simple moving average (SMA)). We consider N values ranging from 2 to 5
days for a total of four different RSI indicators.
2. Stochastic: Developed by George Lane in the 1950s (Schade (2015)), the Stochastic
oscillator is another widely known indicator. Stochastic %K measures how far the
current price is from its N-day low as a percentage of the width of its N-day range.
Stochastic %D takes an M-period SMA of Stochastic %K and Stochastic %DD takes
an L-period SMA of Stochastic %D. That is,
Ct − L N,t
%Kt = 100 ×
HN,t − L N,t
%Dt = SMA(%Kt , M )
%DDt = SMA(%Dt , L)

where HN,t and L N,t are the N-day high and low, respectively, at time t. We consider
%K indicators with N values ranging from 1 to 5 days, all subsequent combinations
of %D indicators with M ranging from 2 to 5 days, and %DD indicators with L equal
to 2. This gives four %K indicators, 21 %D indicators, and 80 %DD indicators. (In our
results tables, we refer to the %K, %D, and %DD indicators as STOK, STOD, STODD,
respectively.)
3. Moving Average Convergence-Divergence (MACD): Another widely followed in-
dicator, MACD takes the K-period exponential moving average (EMA) of the dif-
ference between the M- and N-period EMAs of the closing price. In the original, K,
M, and N were set to 9, 12, and 26 periods, respectively. Here, we look at shorter
horizons ranging from 1 to 5 days and include all combinations that obey K < M <

Electronic copy available at: https://ssrn.com/abstract=4339128

N, which gives us a total of 10 MACD indicators.
Note: The Percentage Price Oscillator (PPO) is a normalized MACD. We use the PPO
rather than the MACD because of this normalization (but refer to it as MACD in our
results tables).
EMA(Ct , M ) − EMA(Ct , N )

PPOt = EMA ,K
EMA(Ct , N )

4. Ultimate Oscillator (UO): The Ultimate Oscillator takes a weighted average of three
quantities, each being the ratio of total “buying pressure” to total “true range” over a
given period. The original formulation by Larry Williams (Wikipedia (2023d)) uses
periods of 7, 14, and 28 days. As with MACD, we use all combinations of periods
ranging from 1 to 5 days with K < M < N below. This yields 10 UO indicators.
K × avgK + M × avg M + N × avg N
UOt = 100 ×
K + M + N
bpt− L+1 + bpt− L+2 + ... + bpt
avg L =
trt− L+1 + trt− L+2 + ... + trt
(Buying Pressure) bpt = Ct − min( Lt , Ct−1 )
(True Range) trt = max( Ht , Ct−1 ) − min( Lt , Ct−1 )

5. Money Flow Index (MFI): The MFI calculates the “money flow” of a period by
multiplying the period’s “typical price” (i.e., the average of the high, low, and close)
by the volume during the period. It then takes the ratio of total positive money flow
to total money flow over a specified period (Wikipedia (2023c)). We consider periods
ranging from 3 to 5 days (which generates three total MFI indicators).
positive money flow
MFIt = 100 ×
positive money flow + negative money flow
money flowt = typical pricet × Vt
Ht + Lt + Ct
typical pricet =
3
positive money flow = sum of money flow where: typical pricet > typical pricet−1
negative money flow = sum of money flow where: typical pricet < typical pricet−1

6. Chaikin Oscillator (CHIOSC): The Chaikin Oscillator, named after its developer,

Electronic copy available at: https://ssrn.com/abstract=4339128

Marc Chaikin (StockCharts (2023a)), takes the difference between N- and M-period
EMAs of the “accumulation-distribution line” (ADL), where ADL is the cumulative
sum of “money flow volume” (MFV) and MFV is the volume times the “money
flow multiplier” (see below). We consider N and M values ranging from 1 to 5 days
(resulting in 10 CHIOSC indicators).

CHIOSCt = EMA(ADLt , N ) − EMA(ADLt , M)

ADLt = ADLt−1 + MFVt
MFVt = MFMt × Vt
[(Ct − Lt ) − ( Ht − Ct )]
MFMt =
Ht − Lt

7. Bollinger Bands™ Index (%B): Bollinger Bands™, named after their progenitor,
John Bollinger (Wikipedia (2023a)), are found by adding and subtracting two stan-
dard deviations of price from an N-period SMA. %B measures the distance the price
is from the lower band as a percentage of the width of the bands. We consider N
values of 2, 3, 4, 5, 10, and 20 for a total of six %B indicators. (We refer to %B as BBI
in our results tables.)
Ct − (µt − 2 × σt )
%Bt = 100 ×
4 × σt
µt = SMA(Ct , N )
σt = N-period standard deviation of Ct

8. Keltner Channel Index (KCI): A Keltner Channel (StockCharts (2023b)) is similar to

Bollinger Bands™, except instead of adding and subtracting the standard deviation
of price from a simple moving average, it adds and subtracts the “average true range”
(ATR) from an exponential moving average. The “true range” is the daily range that
takes into account any gaps from yesterday’s close. We consider EMA lengths and
ATR lengths ranging from 1 to 5 days (for a total of 24 KCI indicators).

10

Electronic copy available at: https://ssrn.com/abstract=4339128

Ct − (µt − 2 × ATRt )
KCIt = 100 ×
4 × ATRt
µt = EMA(Ct , N )
ATRt = SMA(true ranget , N )
true ranget = max( Ht , Ct−1 ) − min( Lt , Ct−1 )

9. Commodity Channel Index (CCI): The CCI, introduced by Donald Lambert in 1980
(Wikipedia (2023b)), takes the difference between the typical price and its N-period
SMA and divides by a quantity equal to a constant multiplied by the mean absolute
deviation of the typical price from its N-period SMA. Following the original formula-
tion, we use 0.015 as the value of the constant. We consider N values ranging from 2
to 5 days (for a total of four CCI indicators).
TPt − SMA(TPt , N )
CCIt =
0.015 × MAD(TPt , N )
Ht + Lt + Ct
TPt =
3
MAD( X, N ) = N-period mean absolute deviation of X from SMA(X,N)

10. True Strength Index (TSI): The TSI is a doubly smoothed indicator created by
William Blau (Blau (1991)). It takes the N-period EMA of the M-period EMA of
the one-period difference in closing prices, then divides by a similar quantity that
uses the absolute difference of closing prices. We consider all combinations of M
values ranging from 2 to 5 days and N values ranging from 1 to 5 days (for a total of
20 TSI indicators).
EMA(EMA(Ct − Ct−1 , M ), N )
TSIt =
EMA(EMA(|Ct − Ct−1 |, M ), N )

11. David Varadi Oscillator (DVO): The DVO is the N-period percent rank of the M-
period simple moving average of the ratio between the period’s close and the mid-
point of the period’s range (Grover (2020)). We consider M values of 1, 2, 3, 4, 5,
and 10 in conjunction with N values of 14, 21, 42, and 63 days. We also consider the
{M, N} pairs of {10, 14} and {14, 14} (giving us a total of 22 DVO indicators).

11

Electronic copy available at: https://ssrn.com/abstract=4339128

! !
Ct
DVOt = PercentRank SMA Ht + Lt
,M ,N
2

12. David Varadi Intermediate Oscillator (DVI): The DVI is a composite indicator that
blends information on return magnitude and “stretch,” i.e., the number of up periods
versus down periods over a given time window (Varadi (2009), Ulrich (2022)2 ).

DVIt = w1 × magnitudet + w2 × stretcht

SMA(rt ,N2 )
! !
N1 × SMA(rt , N1 ) + N2 × 10
magnitudet = PercentRank SMA , N3 , 252
2
SMA(bt ,M2 )
! !
M1 × SMA(bt , M1 ) + M2 × 10
stretch = PercentRank SMA , M3 , 252
2
Ct
rt = −1
SMA(Ct , K )
bt = 1 if Ct > Ct−1 else -1

The original parameters specified by Varadi are:

w1 , w2 = {0.8, 0.2}
K=3
N1 , N2 , N3 = {5, 100, 5}
M1 , M2 , M3 = {10, 100, 2}

We use the original parameters as well as all combinations of the following variations:
SMA(rt ,N2 )

2 SMA(rt ,N1 )+ 10
The literature presents the “magnitude” as being given by PercentRank SMA 2 , N3 , 252 ,
but the SMA term divided by 10 does not appear to make sense. A more sensible formulation would be one
equivalent to the “stretch” calculation, where sums rather than SMAs are used before averaging the result
and smoothing it. We use SMAs multiplied by their period lengths to represent these sums.

12

Electronic copy available at: https://ssrn.com/abstract=4339128

w1 , w2 = {0.5, 0.5}, {0.2, 0.8}
K = {2}, {4}, {5}
N1 , N2 , N3 = {3, 50, 5}, {2, 20, 5}
M1 , M2 , M3 = {7, 70, 2}, {5, 50, 2}

This results in a total of 36 DVI indicators.

In addition to the foregoing indicators, we also investigate the following modified indica-
tors:

1. Modified True Strength Index (MTSI): In this modified TSI, we use the natural
logarithm of the ratio between the close and the one-day volume-weighted average
price (VWAP) (we also consider a version using the time-weighted average price
(TWAP)). (We compute the VWAP using 1-minute data—by averaging the high-low-
close of a one-minute bar, multiplying by its volume, and aggregating over all bars
of the day.) We consider all combinations of M values ranging from 2 to 5 days and
N values ranging from 1 to 5 days (for a total of 40 TSI indicators, when both the
VWAP and TWAP versions are counted).
EMA(EMA(ln(Ct /VWAPt ), M), N )
MTSIt = 100 ×
EMA(EMA(ln(|Ct /VWAPt |), M), N )

2. Modified RSI (MRSI): This modified RSI formulation incorporates information on

the intraday range rather than day-to-day changes in price. It uses Wilder Mov-
ing Averages of the natural logarithms of the high-to-close and close-to-low ratios.
Further, it speeds up or slows down the oscillation by specifying an exponent, N.
UtN
MRSIt = 100 ×
UtN + DtN

Ht
Ut = WMA ln ,M
Ct

Ct
Dt = WMA ln ,M
Lt

We consider all combinations of M values between 2 and 5 and N values between 1

13

Electronic copy available at: https://ssrn.com/abstract=4339128

and 5 (for a total of 20 MRSI indicators).

3. Volume-Weighted Modified RSI (VWMRSI): The VWMRSI weights the high-to-

close and close-to-low ratios described above with a multiplier based on the day’s
volume relative to the SMA of volume over a specified period, K. The underlying
premise is that near-term price extremes are more likely to reverse if they are accom-
panied by higher volume (Campbell et al. (1993)). Thus, the VWMRSI amplifies or
mutes the high-to-close and close-to-low ratios based on the relative level of volume.
We set K = 21 days. As with MRSI, we consider all combinations of M values
between 2 and 5 and N values between 1 and 5 (for a total of 20 VWMRSI indicators).
UtN
VWMRSIt = 100 ×
UtN + DtN

Ht
Ut = WMA V̂ × ln ,M
C
t
Ct
Dt = WMA V̂ × ln ,M
Lt
Vt
V̂ =
SMA(Vt , K )

The author developed the MRSI and MTSI in the context of short-term macro strategies
in the 2009-2010 time frame, as co-founder of the investment firm, MacroSignal Partners.
These indicators formed the basis of several strategies among a larger ensemble that
focused on mean-reversion, momentum, and cross-asset dynamics.

The suite of strategies ran for 32 months from 2011 to 2014 (27 month at MacroSignal and
five months at a larger hedge fund), generating near-double digit annualized returns with
a Sharpe ratio of 1.5 during a time when the HFRX Global Hedge Fund Index was flat and
the HFRX Systematic Diversified Macro Index was down 10%.

This paper in large part grew out of a desire to see how the MTSI and MRSI indicators
have performed out-of-sample as well as to compare them with other prominent price
oscillators.

14

Electronic copy available at: https://ssrn.com/abstract=4339128

II. Strategies and Methodology
Let Xi,A,t represent the value at time t of the ith indicator applied to asset A. As not all
indicators range between 0 and 100, we first convert each indicator into a cumulative
percent rank, Yi,A,t (using a one-year burn-in period):

Yi,A,t = PercentRank( Xi,A,t , cumulative = TRUE)

Further, because some indicators, such as the Chaikin Oscillator, have distributions that
shift over time, the frequency of observations below (or above) a given expanding-window
percent rank could differ markedly from the expected frequency of observations (e.g., 20%
of observations below the cumulative 10th percentile).

To counter this, we find the values, αi,q for each combination of indicator and desired
quantile, q, such that across all assets, the median percentage of indicator values (Yi,A,t )
that fall below αi,q is q%.

We then construct idealized3 long-only and short-only strategies as follows. Starting with
a given (percent rank) indicator, Yi,A,t :

Long :Go long asset, A, at the close at time t and hold for h days if
Yi,A,t < αi,q , (h = 1, 3, 5, q = 10, 20, 30)

Short :Go short asset, A, at the close at time t and hold for h days if
Yi,A,t > αi,q , (h = 1, 3, 5, q = 70, 80, 90)

With three holding periods and three quantiles, there are nine different long and short
strategies.

To assess the performance of each indicator, we do the following: For a given quadruplet
consisting of a direction (long or short), an indicator, a holding period, and a quantile, we
compute the Sharpe ratio of returns for each asset in the universe and record the median
value. We also create an equal-weighted portfolio for each quadruplet by aggregating the
asset-specific strategy returns. We then record the Sharpe ratio of this portfolio. Finally,
3
We assume we know the closing price to compute the indicator and then are able to enter at the closing
price if a signal is given. We also do not consider trading costs or slippage.

15

Electronic copy available at: https://ssrn.com/abstract=4339128

we compute the composite rank of the median asset-specific Sharpe ratio and the portfolio
Sharpe ratio.

As an example, let’s consider a 2-day RSI (RSI(2)) strategy that goes long an asset at the
close and holds for h days when the value of RSI(2) for that asset is below α RSI (2),q . We
apply the strategy to all assets in the universe and compute the asset-specific strategy
returns. We then compute the asset-specific Sharpe ratios and record the median value
across all assets. Lastly, we compound the asset-specific strategy returns and average them
at each point in time to get the portfolio’s total return. We then back out the daily portfolio
returns from this total return and compute the portfolio’s Sharpe ratio.

Next, we do the same for all the other indicators (for a given direction, holding period,
and “true” quantile}. Once completed, we have a median asset-specific Sharpe ratio and
portfolio Sharpe ratio for each indicator. We rank those values, add the ranks, and rank
the sum to arrive at our overall performance metric (for a given direction, holding period,
and quantile).

We further aggregate those performance metrics by taking the composite composite ranks
of all:

1. Strategies (both long and short, all holding periods, all quantiles)
2. Long strategies (all holding periods, all quantiles)
3. Short strategies (all holding periods, all quantiles)
4. 1-day holding period strategies (both long and short, all quantiles)
5. 1-day holding period long strategies (all quantiles)
6. 1-day holding period short strategies (all quantiles)
7. 3-day holding period strategies (both long and short, all quantiles)
8. 3-day holding period long strategies (all quantiles)
9. 3-day holding period short strategies (all quantiles)
10. 5-day holding period strategies (both long and short, all quantiles)
11. 5-day holding period long strategies (all quantiles)
12. 5-day holding period short strategies (all quantiles)

In this manner, we can see which indicators were the best overall, which were the best for
long strategies, short strategies, 1-day holding periods, 3-day holding periods, etc.

For each of the nine long and short strategies, we also compute the highest percent rank
and highest composite Sharpe ratio4 for each indicator family.
4
The composite Sharpe ratio is the average of the median asset-specific Sharpe ratio and portfolio Sharpe
ratio for a given indicator, direction, holding period, and quantile.

16

Electronic copy available at: https://ssrn.com/abstract=4339128

In addition to performing the analysis on the full dataset, we also examine the first half
(2003-2012) and second half (2013-2022) separately. We present those results in Appendices
A and B, respectively.

III. Results
Table I shows, for the full dataset, the percentage composition of the top quintile by
indicator family across the various aggregations (listed in the previous section). MTSI had
the most entries in the top quintile of performance for 11 of the 12 aggregations—typically
representing between 40% and 60% of the top performing indicators.

In our composite measure of overall performance (across directions, holding periods, and
quantiles), MTSI represented 58% of the top performing indicators—garnering the first
30 spots in the top quintile as well as eight additional spots for a total of 38 (with 334
indicators in total, the top quintile consisted of 66 indicators). This represented 95% of all
MTSI indicators considered. (Notably, the MTSI indicators based on TWAP do slightly
better than the ones based on VWAP.)

VWMRSI had the second-largest representation in our measure of overall performance at

11% (seven indicators, 35% of all VWMRSI indicators), followed by CHIOSC, MRSI, and
UO at 9%, 8%, and 8%, respectively. MRSI indicators held spots 31 through 35, right after
MTSI (which held spots 1-30).

Nine of the 17 indicator families (RSI, STODD, MACD, KCR, CCI, MFI, BBI, DVI, and TSI)
had no representation in the top quintile of overall performance.

Notably, STODD, while not registering in the top quintile of overall performance, had
strong representation in the 3- and 5-day holding period aggregations. MRSI, in contrast,
fared particularly well in the 1-day holding period aggregations.

Table VI in Appendix A shows the comparable table for the first half of the dataset. Again,
we see that MTSI tended to dominate the ranks of the top performing indicators—having
the largest representation in six of the 12 aggregations, including 61% representation in the
overall metric. UO, STOD, and MRSI had the next largest representations in the overall
performance rankings at 9%, 9%, and 8%, respectively. As with the full dataset, STODD
performed well in the 3- and 5-day holding period aggregations, while MRSI performed
better with 1-day holding periods.

Table XI in Appendix B shows the corresponding results for the second half of the dataset.

17

Electronic copy available at: https://ssrn.com/abstract=4339128

Yet again, MTSI dominated the field, having the largest representation in nine of the 12
aggregations, including 59% of the overall top quintile. CHIOSC, UO, and MRSI were
the next largest components in the overall top quintile, coming in at 12%, 9%, and 8%
respectively. As before, STODD, while not registering in the overall rankings, did perform
well for 3- and 5-day holding periods. Likewise, MRSI performed relatively well in 1-day
holding period strategies.

Table II shows the maximum percentile of performance achieved by each indicator family
across the nine different long strategies for the full dataset. Table III shows the same
information for the short strategies.

As one would expect from the composite rankings, MTSI again dominated, fielding the
top ranked indicator in 6 of the 9 long strategies, and ranking in the top 2% of the other
3 strategies. On the short side, MTSI was not as dominating, but still achieved top 5%
performance in 6 of the 9 short strategies (and the top performance in two).

Table IV shows the maximum composite Sharpe ratio achieved by each indictor family for
the nine long strategies across the full dataset. Table V shows the corresponding results for
the nine short strategies. Here, MTSI had the highest composite Sharpe in seven of the
nine long strategies and three of the nine short strategies.

Appendices A and B present the maximum percent rank and maximum composite Sharpe
ratio results for the first and second halves of the dataset, respectively. The performance
is as one would expect given the aforementioned top quintile rankings (as the composite
Sharpe ratio was the ranking metric).

IV. Conclusion
Short-term mean-reversion has been an empirical fact of equity markets for the past two
decades. During this time, it has generally been strong enough and persistent enough to
exploit using short-term price oscillators.

In that regard, we investigate an array of price oscillators for their ability to take advantage
of short-term mean-reversion in global equities. The indicators tested include many well-
known, longstanding oscillators as well as several modified oscillators developed about
a decade ago. We build simple 1-, 3-, and 5-day holding period strategies, both long
and short, using normalized indicators (normalized for time in the market). In total, we
consider 334 different indicators arising from 17 different indicator families.

18

Electronic copy available at: https://ssrn.com/abstract=4339128

We find that one of the modified oscillators, the Modified True Strength Index (MTSI),
dominated the performance rankings in the full dataset as well as in both halves sepa-
rately—coming in as the top ranked indicator overall and garnering the top 30 spots in
the overall performance rankings. It did similarly well in nearly all composite aggrega-
tions, individual strategies, and in both halves of the dataset—suggesting a high level of
versatility and robustness.

The two other modified oscillators, the Modified Relative Strength Index (MRSI) and the
Volume-Weighted Modified Relative Strength Index (VWMRSI), also fared well. MRSI
was the second-best performing indicator family in the overall composite rankings on
the full dataset (coming in right behind MTSI) and performed particularly well on 1-day
holding period strategies. VWMRSI had the second-highest percentage representation
in the top quintile of overall performance (for the full dataset) and tended to do well in
3- and 5-day holding period strategies. While MRSI did similarly well in both halves of
the dataset, VWMRSI performed noticeably better in the second half of the dataset. Both
the Ultimate Oscillator and Chaikin Oscillator families also performed well overall across
the full dataset. In contrast, many indicator families failed to place any of their variations
in the top quintile of overall performance. These families included RSI, STODD, MACD,
KCR, CCI, MFI, BBI, DVI, and TSI.

RSI’s relatively poor performance was somewhat surprising given that 2-day RSI is a
popular indicator in many online studies of mean-reversion.

Perhaps the biggest takeaway of the study, however, other than MTSI being the clear
winner, is that the best performing indicators tend to incorporate measures of where the
current price lies in the intraday or multiday trading range (e.g., MTSI, MRSI). In contrast,
the worst performing indicators tend to incorporate day-over-day changes in closing price
(e.g., RSI, TSI).

This tendency may offer a fruitful starting point to future analysts developing their own
indicators.

19

Electronic copy available at: https://ssrn.com/abstract=4339128

Table I. This table presents the percentage representation of each indicator family in the top quintile for each
of the 12 aggregations listed in Section III (2003-2022).

Percentage Representation in Composite Rank Top Quintile by Indicator Family: 2003-2022

All Holding Periods 1-Day Hold 3-Day Hold 5-Day Hold

Family Overall Long Short Overall Long Short Overall Long Short Overall Long Short

RSI 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0%
UO 8% 8% 9% 9% 9% 3% 2% 0% 2% 3% 3% 9%
STOK 3% 0% 5% 3% 3% 3% 0% 0% 2% 6% 2% 6%
STOD 3% 3% 9% 0% 2% 0% 9% 9% 11% 8% 8% 6%
STODD 0% 5% 8% 0% 0% 0% 30% 29% 26% 20% 11% 30%
MACD 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0%
CHIOSC 9% 0% 14% 9% 8% 14% 5% 0% 14% 2% 0% 9%
KCR 0% 0% 2% 0% 0% 0% 0% 0% 0% 0% 0% 0%
CCI 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0%
MFI 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0%
BBI 0% 0% 2% 0% 0% 0% 0% 0% 2% 0% 0% 2%
DVI 0% 0% 0% 0% 0% 0% 0% 0% 2% 0% 0% 3%
DVO 2% 9% 0% 2% 8% 5% 2% 5% 2% 6% 9% 2%
MRSI 8% 8% 8% 23% 23% 24% 0% 0% 0% 2% 0% 8%
VWMRSI 11% 8% 8% 8% 8% 0% 0% 0% 0% 15% 8% 15%
MTSI 58% 61% 38% 47% 41% 52% 53% 58% 42% 39% 61% 11%
TSI 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0%

20

Electronic copy available at: https://ssrn.com/abstract=4339128

Table II. This table presents the maximum percent rank attained by each indicator family in the composite
rankings for each of the nine long strategies (2003-2022).

Maximum Percent Rank by Indicator Family: 2003-2022

Direction: Long

Quantile: 10 20 30

Family Hold: 1-Day 3-Day 5-Day 1-Day 3-Day 5-Day 1-Day 3-Day 5-Day

RSI 58% 42% 36% 55% 22% 70% 60% 33% 65%
UO 99% 89% 81% 96% 91% 85% 88% 67% 85%
STOK 94% 99% 83% 91% 68% 83% 89% 38% 85%
STOD 100% 86% 88% 87% 92% 94% 81% 94% 81%
STODD 71% 96% 88% 60% 100% 90% 62% 94% 94%
MACD 33% 3% 12% 18% 3% 15% 19% 5% 19%
CHIOSC 94% 69% 62% 96% 66% 64% 93% 73% 67%
KCR 74% 66% 65% 62% 52% 63% 59% 41% 70%
CCI 43% 25% 51% 61% 31% 59% 61% 34% 66%
MFI 41% 73% 80% 32% 64% 80% 4% 70% 47%
BBI 53% 57% 75% 60% 38% 60% 63% 28% 79%
DVI 19% 34% 31% 17% 48% 20% 22% 35% 19%
DVO 96% 98% 99% 94% 96% 93% 90% 84% 99%
MRSI 99% 80% 79% 100% 68% 67% 97% 76% 77%
VWMRSI 89% 53% 69% 98% 81% 78% 95% 89% 89%
MTSI 98% 100% 100% 98% 99% 100% 100% 100% 100%
TSI 57% 44% 45% 56% 33% 76% 61% 35% 78%

21

Electronic copy available at: https://ssrn.com/abstract=4339128

Table III. This table presents the maximum percent rank attained by each indicator family in the composite
rankings for each of the nine short strategies (2003-2022).

Maximum Percent Rank by Indicator Family: 2003-2022

Direction: Short

Quantile: 10 20 30

Family Hold: 1-Day 3-Day 5-Day 1-Day 3-Day 5-Day 1-Day 3-Day 5-Day

RSI 54% 59% 54% 60% 70% 37% 60% 53% 35%
UO 93% 89% 96% 85% 83% 96% 87% 71% 83%
STOK 99% 99% 100% 90% 92% 99% 73% 59% 89%
STOD 95% 96% 93% 72% 99% 100% 71% 100% 97%
STODD 77% 90% 96% 64% 100% 99% 66% 99% 100%
MACD 70% 33% 54% 46% 8% 5% 28% 6% 4%
CHIOSC 100% 99% 99% 100% 94% 90% 88% 90% 90%
KCR 87% 86% 81% 66% 81% 78% 65% 66% 69%
CCI 21% 58% 96% 18% 32% 67% 26% 24% 44%
MFI 59% 83% 90% 38% 40% 67% 32% 44% 37%
BBI 23% 92% 75% 65% 98% 94% 70% 92% 93%
DVI 90% 84% 93% 51% 81% 78% 35% 84% 79%
DVO 97% 93% 97% 80% 83% 80% 72% 67% 70%
MRSI 83% 63% 51% 92% 65% 88% 94% 87% 86%
VWMRSI 66% 30% 78% 90% 51% 97% 91% 89% 92%
MTSI 96% 100% 93% 99% 98% 75% 100% 96% 84%
TSI 81% 66% 55% 59% 78% 31% 51% 61% 36%

22

Electronic copy available at: https://ssrn.com/abstract=4339128

Table IV. This table presents the maximum composite Sharpe ratio attained by each indicator family in the
composite rankings for each of the nine long strategies (2003-2022).

Maximum Composite Sharpe by Indicator Family and Case: 2003-2022

Direction: Long

Quantile: 10 20 30

Family Hold: 1-Day 3-Day 5-Day 1-Day 3-Day 5-Day 1-Day 3-Day 5-Day

RSI 1.35 0.74 0.67 1.02 0.66 0.77 0.97 0.70 0.74
UO 1.78 1.00 0.86 1.59 0.91 0.83 1.38 0.80 0.78
STOK 1.74 1.11 0.88 1.52 0.84 0.81 1.39 0.72 0.79
STOD 1.93 0.97 0.96 1.49 0.96 0.91 1.32 0.91 0.78
STODD 1.51 1.06 0.95 1.12 1.02 0.87 1.04 0.91 0.83
MACD 0.97 0.40 0.51 0.73 0.48 0.57 0.72 0.52 0.61
CHIOSC 1.74 0.85 0.77 1.62 0.83 0.76 1.48 0.83 0.74
KCR 1.56 0.84 0.82 1.15 0.75 0.75 0.96 0.72 0.74
CCI 1.21 0.65 0.74 1.13 0.69 0.75 1.00 0.71 0.74
MFI 1.19 0.88 0.85 0.83 0.84 0.79 0.62 0.83 0.72
BBI 1.29 0.81 0.83 1.11 0.73 0.75 1.04 0.69 0.77
DVI 0.81 0.71 0.64 0.74 0.77 0.62 0.75 0.71 0.61
DVO 1.75 1.00 1.12 1.56 0.94 0.90 1.43 0.86 0.90
MRSI 1.76 0.92 0.85 1.64 0.82 0.77 1.52 0.85 0.76
VWMRSI 1.66 0.80 0.80 1.64 0.88 0.78 1.47 0.89 0.81
MTSI 1.80 1.14 1.13 1.67 0.96 1.00 1.62 0.97 0.91
TSI 1.34 0.76 0.72 1.03 0.71 0.78 1.02 0.71 0.77

23

Electronic copy available at: https://ssrn.com/abstract=4339128

Table V. This table presents the maximum composite Sharpe ratio attained by each indicator family in the
composite rankings for each of the nine short strategies (2003-2022).

Maximum Composite Sharpe by Indicator Family and Case: 2003-2022

Direction: Short

Quantile: 10 20 30

Family Hold: 1-Day 3-Day 5-Day 1-Day 3-Day 5-Day 1-Day 3-Day 5-Day

RSI 0.47 −0.07 −0.19 0.38 −0.09 −0.26 0.25 −0.16 −0.31
UO 0.77 0.14 −0.01 0.66 −0.03 −0.12 0.58 −0.12 −0.23
STOK 1.32 0.34 0.13 0.71 0.02 −0.09 0.45 −0.14 −0.21
STOD 0.88 0.22 −0.02 0.55 0.09 −0.01 0.38 0.01 −0.13
STODD 0.63 0.17 0.01 0.44 0.15 −0.05 0.30 0.01 −0.06
MACD 0.59 −0.18 −0.21 0.30 −0.33 −0.46 0.03 −0.38 −0.44
CHIOSC 1.56 0.31 0.18 0.94 0.03 −0.14 0.60 −0.08 −0.19
KCR 0.70 0.12 −0.12 0.45 −0.05 −0.17 0.30 −0.12 −0.25
CCI 0.22 −0.07 0.02 0.06 −0.16 −0.20 0.02 −0.22 −0.29
MFI 0.25 −0.26 −0.34 0.23 −0.14 −0.20 0.06 −0.17 −0.30
BBI 0.23 0.17 −0.13 0.42 0.10 −0.12 0.35 −0.06 −0.18
DVI 0.75 0.09 −0.03 0.33 −0.04 −0.17 0.09 −0.09 −0.24
DVO 1.09 0.20 0.04 0.66 −0.03 −0.16 0.40 −0.12 −0.25
MRSI 0.66 −0.03 −0.22 0.72 −0.10 −0.15 0.64 −0.08 −0.21
VWMRSI 0.54 −0.19 −0.12 0.74 −0.11 −0.11 0.63 −0.08 −0.18
MTSI 0.91 0.36 −0.04 0.88 0.04 −0.17 0.76 −0.03 −0.23
TSI 0.63 −0.03 −0.19 0.37 −0.05 −0.27 0.19 −0.13 −0.31

24

Electronic copy available at: https://ssrn.com/abstract=4339128

Appendix A.
Table VI. This table presents the percentage representation of each indicator family in the top quintile for
each of the 12 aggregations listed in Section III (2003-2012).

Percentage Representation in Composite Rank Top Quintile by Indicator Family: 2003-2012

All Holding Periods 1-Day Hold 3-Day Hold 5-Day Hold

Family Overall Long Short Overall Long Short Overall Long Short Overall Long Short

RSI 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0%
UO 9% 9% 9% 9% 9% 8% 5% 8% 0% 2% 3% 2%
STOK 3% 5% 3% 5% 5% 3% 0% 2% 2% 5% 5% 3%
STOD 9% 3% 14% 0% 2% 2% 11% 8% 14% 14% 9% 9%
STODD 6% 3% 23% 0% 0% 0% 38% 35% 39% 44% 18% 64%
MACD 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0%
CHIOSC 5% 0% 14% 9% 8% 14% 0% 0% 14% 0% 0% 11%
KCR 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0%
CCI 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0%
MFI 0% 0% 0% 0% 0% 0% 0% 0% 0% 2% 2% 0%
BBI 0% 0% 2% 0% 0% 0% 0% 0% 2% 0% 2% 2%
DVI 0% 0% 6% 0% 0% 0% 0% 0% 5% 0% 0% 9%
DVO 0% 5% 0% 2% 8% 0% 2% 8% 0% 2% 5% 2%
MRSI 8% 8% 0% 15% 15% 20% 0% 0% 0% 0% 0% 0%
VWMRSI 0% 8% 0% 8% 15% 0% 0% 3% 0% 5% 0% 0%
MTSI 61% 61% 30% 53% 39% 55% 45% 38% 26% 29% 56% 0%
TSI 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 2% 0%

25

Electronic copy available at: https://ssrn.com/abstract=4339128

Table VII. This table presents the maximum percent rank attained by each indicator family in the composite
rankings for each of the nine long strategies (2003-2012).

Maximum Percent Rank by Indicator Family: 2003-2012

Direction: Long

Quantile: 10 20 30

Family Hold: 1-Day 3-Day 5-Day 1-Day 3-Day 5-Day 1-Day 3-Day 5-Day

RSI 53% 46% 56% 53% 38% 99% 66% 67% 97%
UO 98% 93% 82% 100% 99% 83% 97% 95% 96%
STOK 100% 100% 84% 97% 92% 93% 98% 79% 98%
STOD 77% 89% 96% 86% 93% 96% 87% 98% 93%
STODD 60% 90% 94% 55% 100% 99% 62% 100% 94%
MACD 53% 4% 10% 42% 3% 8% 25% 4% 12%
CHIOSC 97% 62% 25% 85% 34% 26% 93% 39% 30%
KCR 96% 85% 78% 74% 85% 89% 69% 64% 95%
CCI 48% 47% 63% 65% 62% 81% 74% 51% 91%
MFI 71% 80% 96% 29% 75% 96% 13% 73% 92%
BBI 61% 78% 74% 66% 50% 92% 69% 43% 98%
DVI 31% 35% 33% 27% 71% 29% 31% 72% 31%
DVO 95% 99% 98% 95% 98% 57% 96% 93% 70%
MRSI 89% 75% 55% 99% 61% 45% 93% 76% 85%
VWMRSI 99% 72% 66% 94% 78% 59% 95% 85% 90%
MTSI 97% 99% 100% 97% 92% 97% 100% 99% 96%
TSI 56% 64% 89% 56% 49% 100% 66% 71% 100%

26

Electronic copy available at: https://ssrn.com/abstract=4339128

Table VIII. This table presents the maximum percent rank attained by each indicator family in the composite
rankings for each of the nine short strategies (2003-2012).

Maximum Percent Rank by Indicator Family: 2003-2012

Direction: Short

Quantile: 10 20 30

Family Hold: 1-Day 3-Day 5-Day 1-Day 3-Day 5-Day 1-Day 3-Day 5-Day

RSI 61% 39% 34% 61% 71% 47% 65% 56% 56%
UO 97% 67% 76% 92% 71% 73% 92% 75% 75%
STOK 98% 87% 78% 91% 90% 90% 75% 59% 72%
STOD 94% 92% 95% 85% 100% 99% 73% 100% 97%
STODD 81% 96% 97% 68% 99% 100% 64% 99% 100%
MACD 44% 40% 58% 19% 6% 7% 10% 5% 5%
CHIOSC 100% 98% 99% 100% 94% 95% 96% 86% 90%
KCR 90% 65% 79% 68% 75% 65% 68% 77% 70%
CCI 16% 50% 59% 14% 37% 65% 22% 34% 57%
MFI 59% 65% 82% 50% 75% 80% 34% 73% 77%
BBI 21% 69% 51% 57% 92% 63% 67% 82% 75%
DVI 93% 95% 100% 62% 95% 96% 48% 99% 96%
DVO 87% 68% 94% 64% 67% 73% 73% 64% 61%
MRSI 70% 53% 19% 84% 43% 50% 87% 58% 51%
VWMRSI 84% 42% 42% 90% 46% 61% 99% 46% 53%
MTSI 97% 100% 91% 99% 79% 58% 100% 87% 69%
TSI 67% 46% 49% 73% 74% 57% 58% 69% 65%

27

Electronic copy available at: https://ssrn.com/abstract=4339128

Table IX. This table presents the maximum composite Sharpe ratio attained by each indicator family in the
composite rankings for each of the nine long strategies (2003-2012).

Maximum Composite Sharpe by Indicator Family and Case: 2003-2012

Direction: Long

Quantile: 10 20 30

Family Hold: 1-Day 3-Day 5-Day 1-Day 3-Day 5-Day 1-Day 3-Day 5-Day

RSI 1.48 0.83 0.83 1.11 0.73 0.97 1.13 0.81 0.83
UO 2.42 1.15 0.94 2.08 1.09 0.84 1.71 0.89 0.81
STOK 2.50 1.31 0.96 1.95 0.96 0.88 1.76 0.84 0.83
STOD 2.06 1.10 1.08 1.69 1.00 0.91 1.46 0.92 0.80
STODD 1.65 1.14 1.06 1.16 1.14 0.95 1.12 0.99 0.81
MACD 1.47 0.41 0.46 0.98 0.39 0.44 0.79 0.45 0.54
CHIOSC 2.38 0.88 0.67 1.66 0.72 0.66 1.60 0.76 0.66
KCR 2.37 1.05 1.00 1.44 0.93 0.85 1.15 0.81 0.80
CCI 1.37 0.83 0.86 1.33 0.80 0.84 1.20 0.79 0.80
MFI 1.32 1.02 1.01 0.88 0.84 0.91 0.72 0.83 0.80
BBI 1.71 0.99 0.92 1.31 0.78 0.87 1.16 0.77 0.83
DVI 1.11 0.79 0.73 0.86 0.83 0.70 0.83 0.85 0.67
DVO 2.36 1.14 0.97 1.86 1.07 0.78 1.60 0.89 0.77
MRSI 2.19 0.97 0.83 1.93 0.80 0.75 1.58 0.83 0.79
VWMRSI 2.43 0.95 0.87 1.86 0.85 0.78 1.59 0.86 0.79
MTSI 2.37 1.28 1.16 1.94 0.97 0.92 2.01 0.93 0.82
TSI 1.54 0.92 1.03 1.18 0.78 0.97 1.16 0.82 0.86

28

Electronic copy available at: https://ssrn.com/abstract=4339128

Table X. This table presents the maximum composite Sharpe ratio attained by each indicator family in the
composite rankings for each of the nine short strategies (2003-2012).

Maximum Composite Sharpe by Indicator Family and Case: 2003-2012

Direction: Short

Quantile: 10 20 30

Family Hold: 1-Day 3-Day 5-Day 1-Day 3-Day 5-Day 1-Day 3-Day 5-Day

RSI 0.93 0.02 −0.14 0.83 0.07 −0.13 0.65 −0.04 −0.17
UO 1.54 0.20 0.11 1.19 0.06 −0.01 1.01 0.04 −0.05
STOK 1.75 0.37 0.12 1.20 0.20 0.12 0.85 −0.03 −0.07
STOD 1.48 0.47 0.24 1.15 0.37 0.22 0.86 0.26 0.10
STODD 1.18 0.52 0.26 0.94 0.37 0.23 0.66 0.23 0.20
MACD 0.75 0.04 0.02 0.33 −0.22 −0.30 0.11 −0.29 −0.31
CHIOSC 2.18 0.57 0.37 1.53 0.26 0.18 1.07 0.12 0.01
KCR 1.30 0.17 0.14 0.93 0.12 −0.05 0.71 0.05 −0.08
CCI 0.45 0.06 0.02 0.22 −0.02 −0.06 0.29 −0.08 −0.16
MFI 0.75 −0.06 0.04 0.73 0.12 0.05 0.41 0.03 −0.04
BBI 0.54 0.22 −0.03 0.78 0.24 −0.06 0.70 0.08 −0.05
DVI 1.44 0.52 0.41 0.85 0.27 0.17 0.50 0.25 0.08
DVO 1.24 0.20 0.24 0.86 0.08 0.00 0.83 −0.01 −0.13
MRSI 0.99 0.07 −0.24 1.09 −0.02 −0.13 0.97 −0.04 −0.17
VWMRSI 1.19 0.04 −0.11 1.19 −0.01 −0.09 1.11 −0.06 −0.17
MTSI 1.55 0.69 0.22 1.34 0.13 −0.10 1.16 0.11 −0.07
TSI 0.97 0.05 −0.04 0.97 0.09 −0.09 0.60 0.00 −0.11

29

Electronic copy available at: https://ssrn.com/abstract=4339128

Appendix B.
Table XI. This table presents the percentage representation of each indicator family in the top quintile for
each of the 12 aggregations listed in Section III (2013-2022).

Percentage Representation in Composite Rank Top Quintile by Indicator Family: 2013-2022

All Holding Periods 1-Day Hold 3-Day Hold 5-Day Hold

Family Overall Long Short Overall Long Short Overall Long Short Overall Long Short

RSI 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0%
UO 9% 6% 9% 8% 9% 6% 3% 2% 5% 3% 3% 9%
STOK 3% 0% 6% 3% 2% 3% 0% 0% 3% 6% 3% 6%
STOD 2% 3% 3% 0% 2% 0% 11% 8% 8% 6% 11% 5%
STODD 0% 3% 3% 0% 0% 0% 24% 30% 21% 20% 15% 24%
MACD 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0%
CHIOSC 12% 0% 15% 11% 9% 14% 3% 0% 14% 5% 0% 12%
KCR 0% 0% 2% 0% 0% 0% 0% 0% 2% 0% 0% 2%
CCI 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0%
MFI 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0%
BBI 0% 0% 2% 0% 0% 0% 0% 0% 2% 0% 0% 2%
DVI 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0%
DVO 2% 9% 2% 0% 6% 0% 2% 5% 2% 8% 8% 2%
MRSI 8% 11% 8% 23% 23% 29% 0% 0% 0% 0% 0% 8%
VWMRSI 6% 8% 8% 8% 8% 0% 0% 0% 0% 8% 2% 15%
MTSI 59% 61% 44% 48% 42% 48% 58% 56% 45% 45% 59% 17%
TSI 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0%
rmica425 G9rv29hQ!

30

Electronic copy available at: https://ssrn.com/abstract=4339128

Table XII. This table presents the maximum percent rank attained by each indicator family in the composite
rankings for each of the nine long strategies (2013-2022).

Maximum Percent Rank by Indicator Family: 2013-2022

Direction: Long

Quantile: 10 20 30

Family Hold: 1-Day 3-Day 5-Day 1-Day 3-Day 5-Day 1-Day 3-Day 5-Day

RSI 85% 46% 19% 53% 31% 42% 55% 30% 29%
UO 92% 68% 71% 93% 74% 83% 66% 54% 70%
STOK 60% 77% 86% 76% 41% 59% 60% 38% 68%
STOD 99% 93% 84% 99% 100% 90% 96% 85% 81%
STODD 97% 98% 90% 75% 99% 92% 83% 91% 89%
MACD 14% 21% 62% 18% 33% 59% 22% 34% 63%
CHIOSC 98% 99% 94% 100% 98% 87% 99% 96% 86%
KCR 64% 46% 47% 43% 29% 44% 54% 43% 51%
CCI 49% 14% 37% 45% 28% 47% 48% 39% 50%
MFI 82% 51% 60% 41% 48% 41% 28% 50% 39%
BBI 71% 49% 53% 56% 46% 56% 52% 44% 45%
DVI 21% 48% 46% 31% 48% 34% 33% 43% 38%
DVO 100% 100% 99% 99% 99% 100% 99% 95% 98%
MRSI 94% 85% 79% 97% 73% 77% 98% 79% 73%
VWMRSI 49% 45% 66% 88% 63% 73% 91% 83% 80%
MTSI 89% 97% 100% 98% 99% 99% 100% 100% 100%
TSI 87% 59% 20% 57% 36% 43% 56% 41% 44%

31

Electronic copy available at: https://ssrn.com/abstract=4339128

Table XIII. This table presents the maximum percent rank attained by each indicator family in the composite
rankings for each of the nine short strategies (2013-2022).

Maximum Percent Rank by Indicator Family: 2013-2022

Direction: Short

Quantile: 10 20 30

Family Hold: 1-Day 3-Day 5-Day 1-Day 3-Day 5-Day 1-Day 3-Day 5-Day

RSI 85% 92% 66% 61% 56% 45% 36% 41% 41%
UO 84% 92% 98% 94% 95% 93% 95% 73% 79%
STOK 99% 99% 100% 99% 95% 93% 82% 64% 76%
STOD 92% 87% 74% 56% 80% 62% 64% 82% 84%
STODD 86% 61% 82% 58% 75% 88% 64% 76% 92%
MACD 97% 53% 49% 89% 31% 32% 69% 32% 33%
CHIOSC 98% 91% 89% 99% 84% 84% 92% 77% 78%
KCR 93% 99% 89% 79% 93% 83% 69% 71% 64%
CCI 47% 87% 98% 59% 80% 82% 57% 66% 72%
MFI 75% 91% 82% 42% 62% 60% 26% 49% 55%
BBI 72% 97% 86% 61% 82% 83% 78% 70% 65%
DVI 49% 31% 36% 34% 28% 40% 42% 24% 43%
DVO 100% 100% 99% 100% 99% 98% 99% 89% 81%
MRSI 88% 60% 90% 92% 79% 96% 91% 86% 95%
VWMRSI 91% 81% 97% 85% 77% 100% 84% 93% 100%
MTSI 60% 88% 84% 99% 100% 90% 100% 100% 89%
TSI 85% 93% 65% 62% 58% 44% 43% 54% 44%

32

Electronic copy available at: https://ssrn.com/abstract=4339128

Table XIV. This table presents the maximum composite Sharpe ratio attained by each indicator family in the
composite rankings for each of the nine long strategies (2013-2022).

Maximum Composite Sharpe by Indicator Family and Case: 2013-2022

Direction: Long

Quantile: 10 20 30

Family Hold: 1-Day 3-Day 5-Day 1-Day 3-Day 5-Day 1-Day 3-Day 5-Day

RSI 1.37 0.69 0.47 0.98 0.64 0.64 0.92 0.61 0.64
UO 1.50 0.83 0.82 1.36 0.91 0.88 1.08 0.78 0.75
STOK 1.10 0.96 0.96 1.23 0.71 0.76 1.01 0.67 0.74
STOD 1.80 1.10 0.97 1.50 1.13 0.97 1.38 0.94 0.79
STODD 1.64 1.26 1.02 1.20 1.18 0.97 1.27 1.00 0.89
MACD 0.58 0.51 0.75 0.63 0.64 0.76 0.65 0.63 0.72
CHIOSC 1.72 1.32 1.06 1.56 1.06 0.92 1.40 1.04 0.83
KCR 1.16 0.66 0.64 0.87 0.62 0.66 0.90 0.70 0.71
CCI 0.98 0.42 0.60 0.90 0.61 0.67 0.85 0.65 0.67
MFI 1.07 0.55 0.70 0.84 0.78 0.64 0.74 0.75 0.65
BBI 1.19 0.68 0.69 1.02 0.76 0.74 0.90 0.68 0.66
DVI 0.66 0.69 0.64 0.77 0.75 0.61 0.76 0.69 0.64
DVO 1.88 1.27 1.26 1.50 1.13 1.07 1.38 1.02 0.97
MRSI 1.58 1.00 0.88 1.39 0.90 0.81 1.36 0.90 0.76
VWMRSI 0.95 0.67 0.76 1.29 0.85 0.79 1.34 0.92 0.78
MTSI 1.45 1.18 1.30 1.41 1.07 1.09 1.39 1.17 1.02
TSI 1.39 0.79 0.49 1.05 0.67 0.64 0.95 0.66 0.66

33

Electronic copy available at: https://ssrn.com/abstract=4339128

Table XV. This table presents the maximum composite Sharpe ratio attained by each indicator family in the
composite rankings for each of the nine short strategies (2013-2022).

Maximum Composite Sharpe by Indicator Family and Case: 2013-2022

Direction: Short

Quantile: 10 20 30

Family Hold: 1-Day 3-Day 5-Day 1-Day 3-Day 5-Day 1-Day 3-Day 5-Day

RSI 0.32 0.05 −0.39 −0.03 −0.30 −0.49 −0.35 −0.42 −0.51
UO 0.32 0.03 −0.08 0.30 −0.04 −0.18 0.21 −0.26 −0.36
STOK 0.80 0.21 0.17 0.43 −0.06 −0.19 0.10 −0.30 −0.37
STOD 0.39 −0.07 −0.33 −0.11 −0.18 −0.37 −0.09 −0.21 −0.33
STODD 0.33 −0.25 −0.28 −0.07 −0.20 −0.25 −0.10 −0.22 −0.28
MACD 0.70 −0.32 −0.45 0.25 −0.43 −0.56 −0.01 −0.46 −0.54
CHIOSC 0.76 −0.02 −0.17 0.40 −0.15 −0.27 0.24 −0.22 −0.36
KCR 0.46 0.28 −0.19 0.17 −0.07 −0.27 −0.03 −0.25 −0.43
CCI −0.07 −0.07 −0.05 −0.06 −0.18 −0.28 −0.16 −0.29 −0.39
MFI −0.18 −0.45 −0.82 −0.23 −0.27 −0.41 −0.42 −0.38 −0.47
BBI 0.23 0.16 −0.23 −0.04 −0.16 −0.28 0.06 −0.28 −0.42
DVI −0.09 −0.45 −0.55 −0.27 −0.46 −0.53 −0.31 −0.51 −0.51
DVO 1.03 0.25 0.03 0.62 0.08 −0.16 0.38 −0.14 −0.35
MRSI 0.34 −0.27 −0.16 0.27 −0.20 −0.16 0.19 −0.18 −0.27
VWMRSI 0.35 −0.09 −0.07 0.20 −0.21 −0.14 0.10 −0.13 −0.22
MTSI 0.04 −0.07 −0.27 0.36 0.05 −0.24 0.37 0.06 −0.30
TSI 0.32 0.05 −0.39 −0.03 −0.29 −0.49 −0.28 −0.35 −0.50

REFERENCES
Black, Fischer, 1986, Noise, The Journal of Finance 41, 529–543.

Blau, William, 1991, True strength index, Technical Analysis of Stocks & Commodities 11,
438–446.

Campbell, John Y., Sanford J. Grossman, and Jiang Wang, 1993, Trading volume and serial
correlation in stock returns, Quarterly Journal of Economics 108, 905–939.

Da, Zhi, Qianqiu Liu, and Ernst Schaumburg, 2014, A closer look at the short-term return
reversal, Management Science 60, 658–674.

34

Electronic copy available at: https://ssrn.com/abstract=4339128

French, Kenneth R., 2022, Data library, http://mba.tuck.dartmouth.edu/pages/faculty/
ken.french/data_library.html.

Grover, Alex, 2020, Varadi oscillator, https://www.tradingview.com/script/m5lVzOkh-

Varadi-Oscillator/.

Lo, Andrew W., and A. Craig Mackinlay, 1988, The size and power of the variance ratio
test in finite samples: a monte carlo investigation, NBER Working Paper No. t0066 .

Schade, George A., 2015, Origins of the stochastic oscillator, cmtassociation.org/kb/origins-

of-the-stochastic-oscillator-article/.

Shiller, Robert J., 1984, Stock prices and social dynamics, Brookings Papers on Economic
Activity 12, 457–498.

SqueezeMetrics, 2017, Gamma exposure (gex), https://squeezemetrics.com/monitor/

download/pdf/white_paper.pdf.

Stiglitz, Joseph E., 1989, Using tax policy to curb speculative trading, Journal of Financial
Services 3, 101–115.

StockCharts, 2023a, Chaikin oscillator, https://school.stockcharts.com/doku.php?id=

technical_indicators:chaikin_oscillator.

StockCharts, 2023b, Keltner channels, https://school.stockcharts.com/doku.php?id=

technical_indicators:keltner_channels.

Subrahmanyam, Avanidhar, 2005, Distinguishing between rationales for short-horizon

predictability of stock returns, The Financial Review 40, 11–35.

Summers, Lawrence H., and Victoria P. Summers, 1989, When financial markets work
too well: A cautious case for a securities transactions tax, Journal of Financial Services 3,
261–286.

Ulrich, Joshua, 2022, Package ttr, https://cran.r-project.org/web/packages/TTR/TTR.

pdf.

Varadi, David, 2009, What is the dvi?, https://cssanalytics.wordpress.com/2009/12/13/

what-is-the-dvi/.

Wikipedia, 2023a, Bollinger bands, https://en.wikipedia.org/wiki/Bollinger_Bands.

35

Electronic copy available at: https://ssrn.com/abstract=4339128

Wikipedia, 2023b, Commodity channel index, https://en.wikipedia.org/wiki/
Commodity_channel_index.

Wikipedia, 2023c, Money flow index, en.wikipedia.org/wiki/Money_flow_index.

Wikipedia, 2023d, Ultimate oscillator, en.wikipedia.org/wiki/Ultimate_oscillator.

Wilder, J. Welles, 1978, New Concepts in Technical Trading (Trend Research).

36

Electronic copy available at: https://ssrn.com/abstract=4339128