What J-FGI is, and why I built it
J-FGI stands for Japan Fear & Greed Index, a Japanese equity sentiment index I built for this research. It combines Trend, Market Strength, trading-value Breadth and Realized Stress into a score from 0 to 100. J-Quants supplies the underlying Japanese market data; the selection of dimensions and construction of the score are my own.
When we say that a market is frightened, what have we actually observed?
Prices might be falling. An index might be holding up while more of its underlying securities weaken. Volatility might have risen sharply, or losses might be spreading across previously resilient parts of the market. Each observation describes something real. None gives us a direct reading of investors' collective emotions.
A market can be volatile while rising. A falling index can reflect a few large constituents. Broad participation can accompany both enthusiasm and a mechanical recovery. Volatility, trend and breadth are different views of market behaviour, each with its own ambiguities.
That leaves a useful research question: can several observable dimensions of the Japanese equity market be combined into a coherent description of its risk sentiment?
J-FGI is a research indicator, rather than an official sentiment index published by a Japanese exchange or another institution. The name draws on familiar fear-and-greed language, but it does not make the score interchangeable with other indices.
CNN's Fear & Greed Index combines seven US market indicators. This first J-FGI uses four Japanese market dimensions, without options, credit-market or safe-haven-demand inputs. It does not use CNN scores.
| Indicator | Market and source | Role in this study |
|---|---|---|
| J-FGI | My research construction; four equally weighted Japanese market dimensions | Describe market pressure and examine subsequent returns and risk |
| CNN Fear & Greed Index | Published by CNN; seven US market indicators, including options, credit and safe-haven demand | Conceptual reference; CNN scores are not study inputs |
| Broad Risk-On | Market-state indicator in my personal trading system | Examine overlap with the system's existing state information |
Figure 1. J-FGI and TOPIX through the complete Core observation window, 31 July 2014–30 September 2020. TOPIX is normalized to 100 at the first observation and has its own axis. The 20 and 80 boundaries are descriptive labels; the figure is not a trading-performance chart.
Build the index before looking at the answer
The first Core combined four economic dimensions.
Trend described the broad market's direction relative to its recent history. Market Strength examined the spread of new price highs and lows across eligible listed securities. Breadth compared the trading-value participation of advancing and declining securities. Realized Stress described volatility the broad market had actually experienced.
These dimensions were translated into a common relative historical scale, with higher scores representing the more relaxed or risk-seeking end of the construction. Each dimension has a 25% weight. We fixed that construction before examining its relationship with subsequent outcomes.
The definition stayed fixed throughout the analysis. We wanted to find out what a simple four-dimensional construction actually measured in Japan.
The research reused existing Japanese market data and calculations, following the KabuForge research framework for recording definitions, preserving evidence and checking results.
A 0–100 number is a representation, not a fact about the market
The score is convenient to read. It is not a natural unit of emotion, and a value of 10 does not mean that the market contains twice as much fear as a value of 20.
For display, we used five ranges:
| J-FGI range | Descriptive label |
|---|---|
| 0 ≤ score < 20 | Extreme Fear |
| 20 ≤ score < 40 | Fear |
| 40 ≤ score < 60 | Neutral |
| 60 ≤ score < 80 | Greed |
| 80 ≤ score ≤ 100 | Extreme Greed |
The boundaries organize a continuous variable. They do not establish discontinuities in market behaviour. The follow-up entry experiment separately uses an inclusive score of 20 or lower; a display band and an entry predicate need not have identical boundary conventions.
Low scores describe greater pressure in prices, participation and volatility; high scores describe a calmer market. Whether the score helps explain subsequent returns is the next question.
The obvious hypothesis: does fear predict higher future returns?
The contrarian intuition is attractive. If fear reflects excessive pessimism, perhaps a low J-FGI should precede higher returns. That would give the score an immediate investment interpretation.
The frozen historical sample offered little support for that interpretation. Across 1,505 complete Core observations, the correlation between J-FGI and subsequent 20-trading-day TOPIX returns was approximately −0.065. A descriptive block-bootstrap interval ran from approximately −0.242 to +0.082, crossing zero.
That interval uses blocks to acknowledge persistence and overlapping observations. It does not turn this inspected historical sample into an independent validation exercise. Its practical message is modest: the observed return relationship was weak and uncertain.
Changes in the score were similarly unconvincing as directional signals. The five-trading-day change had a correlation of approximately −0.002 with subsequent 20-trading-day returns; the ten-trading-day change was approximately −0.028. A change in representation did not reveal an obvious return relationship missing from the level.
Fear did not automatically become a contrarian return signal.
The stronger relationship appeared in risk, not return
The risk results looked different.
J-FGI had a historical correlation of approximately −0.492 with subsequent 20-trading-day realized volatility. Its descriptive block-bootstrap interval was approximately −0.622 to −0.295. Higher levels of the score were associated with calmer subsequent volatility in this sample.
The relationship with subsequent maximum drawdown was approximately +0.286, with a descriptive interval of approximately +0.101 to +0.400.
The sign needs explanation. Maximum drawdown here is expressed as a negative return. A drawdown of −5% is closer to zero than a drawdown of −20%. A positive correlation therefore means that higher J-FGI levels were historically associated with shallower subsequent losses.
Figure 2. Continuous J-FGI level and subsequent 20-trading-day outcomes. Points show Pearson correlations; bars show the existing descriptive 95% block-bootstrap intervals. Drawdown is signed negatively, so a positive association corresponds to shallower losses.
Risk conditions tend to persist. A construction containing realized stress and broad-market participation may recognize an environment that continues for a while. That is a plausible interpretation of the evidence, rather than a separate causal finding.
Perhaps fear and greed is telling us what kind of market we are currently in more clearly than where prices will go next.
Did we discover a new signal—or rename an old regime?
My personal trading system already contained Broad Risk-On, an indicator of a broad willingness to take market risk. The system calls such an environment a regime, or market state. Broad Risk-On is my own system indicator, rather than an externally published sentiment index. Here it provides a comparison with an existing research method.
Its ingredients are not identical to J-FGI, but both use related information about trend, participation and risk. Comparing them helps identify how much information the new index adds.
Their contemporaneous correlation across the same complete Core sample was approximately 0.916. That is a high correlation. It does not mean that the two measures move together "92% of the way," nor does it establish that one independently confirms the other.
Figure 3. A high contemporaneous correlation raises a redundancy question. The two constructions use related market information and somewhat different securities eligibility conventions; this comparison is not independent validation.
The appropriate question is whether J-FGI adds anything after accounting for the existing state measure. A descriptive linear residual comparison left only a very small relationship with 20-trading-day returns, approximately +0.043. A limited relationship with subsequent volatility remained, approximately −0.168.
The result is "largely overlapping, but not identical." J-FGI makes market risk easier to read, but most of its information already appears in my existing market-state framework. The economic value of the remaining differences needs further study.
This connects with Why ArcaViso: Reason as Our Guide amid Uncertainty: the usefulness of a new indicator depends on the information it brings.
Four components do not mean four independent pieces of information
The same redundancy question applies inside the composite.
Trend and Market Strength—the high/low component—had a correlation of approximately 0.812. Breadth and Trend had a correlation of approximately 0.564. Realized Stress overlapped less with the other components, with correlations ranging from approximately 0.281 to 0.486.
Figure 4. Pairwise Pearson correlations of the frozen component scores. Pairwise available observations are used, so the cells need not share an identical sample. Market Strength denotes the high/low component.
The Trend–Market Strength relationship has an intuitive explanation: a rising broad market will often accompany more securities reaching new highs. Combining the two can give that related behaviour more representation without adding a wholly separate fact.
Breadth and realized stress offer different views. Removing components one at a time helps identify repeated information; this article retains the original four-component definition.
A composite does not become more informative simply because it contains more components.
What the current J-FGI does not establish
The study began with roughly eighteen years of local market assets. That does not mean we have eighteen years of complete, historically reconstructible J-FGI values.
The complete four-component window contains 1,505 observations between 31 July 2014 and 30 September 2020. Long warm-up requirements and a calendar issue limit that coverage. The original business-day grid retains a day on which the exchange was entirely shut, interrupting the complete windows used by the frozen calculation. We preserved that first construction and documented the issue instead of silently changing it after inspecting outcomes.
The inputs are current-vintage historical records. We have not verified when each record was originally published and available. This study therefore cannot yet reconstruct the complete information an investor could actually have seen then. A date on a record does not establish availability of the same version at that time.
The fact that an index can be calculated historically does not mean that the same historical value could have been reconstructed in real time. The distinction is developed in Point-in-Time Data Is Not a Boolean and the related Point-in-Time Data topic.
The securities pool is a dated listed-security proxy, including eligible ETFs, REITs and foreign ordinary securities. It has not been accepted as a complete historical ordinary-equity universe. Market availability, eligibility and the execution-day calendar require further work.
Future outcome windows overlap. The historical period has been inspected, and no live or independent forward validation is supplied here. External sentiment measures and Japan-native positioning data have not entered the frozen Core. These questions remain for the next review of the data and calculation process.
So what does fear and greed mean here?
J-FGI currently describes how stressed or relaxed the Japanese equity market is more clearly than its next return. I therefore prefer to use it as a scale for reading the market environment.
It also leaves the popular trading intuition unresolved. A weak relationship with a fixed 20-trading-day return does not answer what happens when someone buys during extreme fear and waits, for however long it takes, until extreme greed.
That is the next experiment: Can You Buy Fear and Sell Greed in Japan?
Construction and sources
- Original research: I constructed J-FGI's four components, equal-weight composite and relative historical scale, and calculated the statistics here. Broad Risk-On belongs to my personal trading system and serves only as an information-overlap comparison.
- Market data: Calculations use locally saved J-Quants historical prices, trading values, listed-security information, calendars and TOPIX data. See JPX J-Quants and the official API client and documentation. These describe the data service; the study's version and historical-availability limits are discussed above.
- Conceptual reference: CNN Fear & Greed Index illustrates combining market behaviour into a sentiment score. This study does not reproduce its seven-indicator construction and has no CNN endorsement or validation.