Adding options and margin trading clearly changed the rhythm of buying fear and selling greed: fewer trades, longer holding periods, higher average returns across completed cycles, and deeper worst drawdowns during those positions. Below, we examine these changes through the actual purchase and sale dates.
This time, we connected TOPIX options to the TOPIX branch and Nikkei 225 options to the Nikkei branch represented by the 1321 ETF, with each branch generating its own signals. The other four price-and-volume components and weekly margin-trading data continue to describe the broader Japanese market.
How the comparison works
J-FGI was built by ArcaViso, using underlying data from JPX J-Quants. See KabuForge for background on the research tools. Here, “credit” refers to margin purchases and short sales in Japanese equities.
Each branch compares four components, four components plus matching-index options, four components plus margin trading, and six components. The six are the original four components, one options indicator, and one margin-trading indicator, equally weighted within each model. All four models use the same dates on which every model has a complete score.
The primary rule remains a buy trigger at ≤20 and a sell trigger at ≥80, executed at the close of the next actual trading day after the signal. The options-volume indicator uses a 5-trading-day put/call ratio. Normalization uses the preceding 1260 complete daily observations, with an additional lag of 1 trading day. Weekly margin balances use the preceding 260 distinct weekly observations and enter the model on the 3rd actual trading day after the observation date; they remain usable through the 10th trading day after that inclusion date. When scores are missing, open positions are reported separately as censored.
Nikkei: higher returns, longer holding periods
The Nikkei branch uses the Nikkei 225 options-volume ratio, with the 1321 ETF as the price-performance proxy. The four models share 2232 valid dates, spanning March 29, 2017 to May 22, 2026. Results under the primary rule are as follows:
| Model | Actual entries | Completed／censored | Average return | Average per-trade drawdown | Worst per-trade drawdown | Median holding period (trading days) |
|---|---|---|---|---|---|---|
| Four components | 7 | 6／1 | +5.43% | -16.57% | -26.12% | 169 |
| + Options-volume ratio | 5 | 4／1 | +15.87% | -21.17% | -26.12% | 258.5 |
| + Margin trading | 2 | 2／0 | +18.29% | -21.07% | -31.52% | 442 |
| Six components | 2 | 2／0 | +23.17% | -21.26% | -31.52% | 479 |
Figure 1 | Nikkei primary specification on common score dates: completed-cycle returns and worst per-trade drawdown. Source: ArcaViso calculations and charts; underlying data: JPX J-Quants.
Moving from four to six components reduces completed cycles from 6 to 2 and raises average returns by 17.74 percentage points. However, the median holding period rises from 169 to 479 trading days, and the worst per-trade drawdown deepens from −26.12% to −31.52%. Figure 1 compares averages for different sets of trades; Figure 2 shows how those trades changed.
The margin-trading dimension has a particularly clear effect on cycle counts: adding it alone reduces completed cycles from 6 to 2 and raises the median holding period to 442 trading days. Adding options as well further changes the six-component model’s average return and holding period.
How many original purchase and sale dates remain?
Using four components as the baseline, the table below counts which purchase and sale dates remain, which disappear, and which are new. Earlier or later trades and merged cycles all affect these counts.
| Added dimension | Same-date entries | Original entry dates no longer present | New entry dates | Same-date exits | Original exit dates no longer present | New exit dates |
|---|---|---|---|---|---|---|
| + Options-volume ratio | 3 | 4 | 2 | 0 | 6 | 4 |
| + Margin trading | 0 | 7 | 2 | 1 | 5 | 1 |
| Six components | 1 | 6 | 1 | 0 | 6 | 2 |
Figure 2 | Each holding interval comes from the trade records, with completed and censored intervals shown separately. Source: ArcaViso calculations and charts; underlying data: JPX J-Quants.
The four-component model has 7 actual entries: 6 completed cycles and 1 censored position. The six-component model has only 2 entries. Of the original 7 entry dates, only 1 remains unchanged; 6 no longer appear on their original dates, and 1 new entry date appears. The timeline shows directly how the added dimensions turn several shorter positions into longer ones.
Nikkei trade records:
| Model | Purchase date | Sale date | Holding period (trading days) | Return | Drawdown | MAE | MFE | Status |
|---|---|---|---|---|---|---|---|---|
| Four components | 2018-02-07 | 2019-10-30 | 420 | +5.71% | -20.97% | -11.38% | +12.14% | Completed |
| Four components | 2020-02-26 | 2020-09-15 | 136 | +3.76% | -26.12% | -26.12% | +4.33% | Completed |
| Four components | 2021-05-14 | 2021-09-14 | 84 | +8.14% | -9.48% | -4.91% | +8.14% | Completed |
| Four components | 2021-10-05 | 2022-11-25 | 279 | +1.78% | -16.94% | -11.04% | +7.11% | Completed |
| Four components | 2023-10-23 | 2024-01-10 | 51 | +11.30% | -3.99% | -1.40% | +11.30% | Completed |
| Four components | 2024-07-26 | 2025-05-28 | 202 | +1.88% | -21.91% | -15.87% | +7.73% | Completed |
| Four components | 2026-03-24 | — | — | — | — | — | — | Censored at index-score gap |
| + Options-volume ratio | 2018-02-07 | 2019-11-08 | 426 | +8.18% | -20.97% | -11.38% | +12.14% | Completed |
| + Options-volume ratio | 2020-02-26 | 2021-02-16 | 237 | +35.60% | -26.12% | -26.12% | +35.60% | Completed |
| + Options-volume ratio | 2021-12-01 | 2023-01-25 | 280 | -1.67% | -15.68% | -11.32% | +5.17% | Completed |
| + Options-volume ratio | 2024-08-05 | 2025-05-30 | 198 | +21.38% | -21.91% | -0.55% | +27.35% | Completed |
| + Options-volume ratio | 2026-03-24 | — | — | — | — | — | — | Censored at index-score gap |
| + Margin trading | 2018-02-13 | 2020-12-30 | 701 | +29.76% | -31.52% | -21.79% | +30.08% | Completed |
| + Margin trading | 2022-02-25 | 2022-11-25 | 183 | +6.82% | -10.62% | -6.63% | +9.32% | Completed |
| Six components | 2018-02-07 | 2021-02-16 | 734 | +40.96% | -31.52% | -23.20% | +40.96% | Completed |
| Six components | 2022-02-24 | 2023-01-25 | 224 | +5.38% | -11.00% | -4.97% | +11.27% | Completed |
TOPIX: why the volume specification is unavailable
The TOPIX branch does use TOPIX options; it no longer borrows Nikkei options. However, in the local sample, the 5-day sum of TOPIX call-option volume is zero on 646 dates, making the put/call ratio undefined. The longest uninterrupted interval of valid ratios is 885 trading days, shorter than the 1260 historical observations plus the current observation required by the original rule. Consequently, the TOPIX volume primary specification produces no valid complete six-component scores, and there are no dates common to all four models.
The TOPIX volume ratio lacks enough consecutive valid history to complete the comparison under the existing rules. Below, we instead examine the previously listed open-interest-ratio secondary specification.
The previously specified open-interest-ratio secondary specification
The previously listed alternative uses the TOPIX options put/call open-interest ratio, leaving all other definitions unchanged. This secondary specification has 2232 valid dates common to all four models. Figures 3 and 4 show the results.
| Model | Actual entries | Completed／censored | Average return | Average per-trade drawdown | Worst per-trade drawdown | Median holding period (trading days) |
|---|---|---|---|---|---|---|
| Four components | 7 | 6／1 | +4.21% | -14.68% | -23.03% | 169 |
| + Options open-interest ratio | 3 | 3／0 | +12.87% | -16.46% | -32.22% | 299 |
| + Margin trading | 2 | 2／0 | +6.34% | -20.45% | -32.22% | 442 |
| Six components (open-interest ratio) | 1 | 1／0 | +23.20% | -32.22% | -32.22% | 871 |
Figure 3 | TOPIX open-interest-ratio secondary specification: returns and worst per-trade drawdown. Source: ArcaViso calculations and charts; underlying data: JPX J-Quants.
The four-component model has 6 completed cycles, while the six-component open-interest version has only 1: a purchase on February 14, 2018 and a sale on September 13, 2021, held for 871 trading days, with a return of +23.20%, a maximum drawdown of −32.22%, and a maximum adverse excursion from the purchase price of −27.39%.
That single-cycle return is much higher than the four-component model’s average return of 4.21%; the worst drawdown also deepens from −23.03% to −32.22%. This trade spans the market shock of 2020. The change is concentrated mainly in the long position following the 2018 entry.
| Added dimension | Same-date entries | Original entry dates no longer present | New entry dates | Same-date exits | Original exit dates no longer present | New exit dates |
|---|---|---|---|---|---|---|
| + Options open-interest ratio | 1 | 6 | 2 | 0 | 6 | 3 |
| + Margin trading | 0 | 7 | 2 | 1 | 5 | 1 |
| Six components (open-interest ratio) | 0 | 7 | 1 | 0 | 6 | 1 |
Figure 4 | Trade-by-trade holding intervals for the four models in the TOPIX open-interest secondary specification. Source: ArcaViso calculations and charts; underlying data: JPX J-Quants.
TOPIX trade records:
| Model | Purchase date | Sale date | Holding period (trading days) | Return | Drawdown | MAE | MFE | Status |
|---|---|---|---|---|---|---|---|---|
| Four components | 2018-02-07 | 2019-10-30 | 420 | -4.80% | -22.39% | -19.11% | +4.24% | Completed |
| Four components | 2020-02-26 | 2020-09-15 | 136 | +2.16% | -23.03% | -23.03% | +2.80% | Completed |
| Four components | 2021-05-14 | 2021-09-14 | 84 | +12.50% | -4.82% | -0.24% | +12.50% | Completed |
| Four components | 2021-10-05 | 2022-11-25 | 279 | +3.61% | -14.43% | -9.70% | +5.54% | Completed |
| Four components | 2023-10-23 | 2024-01-10 | 51 | +9.19% | -3.10% | -0.65% | +9.19% | Completed |
| Four components | 2024-07-26 | 2025-05-28 | 202 | +2.59% | -20.30% | -17.50% | +4.29% | Completed |
| Four components | 2026-03-24 | — | — | — | — | — | — | Censored at index-score gap |
| + Options open-interest ratio | 2018-02-07 | 2021-09-15 | 877 | +19.80% | -32.22% | -29.35% | +21.08% | Completed |
| + Options open-interest ratio | 2021-12-14 | 2023-03-07 | 299 | +3.61% | -13.75% | -10.89% | +3.61% | Completed |
| + Options open-interest ratio | 2025-04-08 | 2025-05-30 | 35 | +15.20% | -3.40% | -3.40% | +15.62% | Completed |
| + Margin trading | 2018-02-13 | 2020-12-30 | 701 | +5.12% | -32.22% | -27.98% | +6.25% | Completed |
| + Margin trading | 2022-02-25 | 2022-11-25 | 183 | +7.56% | -8.67% | -6.25% | +7.60% | Completed |
| Six components (open-interest ratio) | 2018-02-14 | 2021-09-13 | 871 | +23.20% | -32.22% | -27.39% | +23.20% | Completed |
What this recalculation shows
Adding matching-index options and margin trading substantially changes buying and selling behavior. The Nikkei primary specification shifts from more short cycles to fewer long cycles. The TOPIX open-interest secondary specification goes further, leaving just one completed cycle spanning several years. Average completed-cycle returns rise, while the worst drawdown during a position deepens and holding periods lengthen.
The added dimensions produce fewer trades, longer positions, higher average returns per completed cycle, and deeper drawdowns along the way. The next question to observe is whether this rhythm persists in a new market phase.
The research materials also retain the 20→70, 30→80, and 30→70 specifications, as well as a specification that waits within the fear zone for the 5-trading-day score change to turn positive. They also include fixed holding periods, historical entry points with matching holding periods, and Buy & Hold comparisons.
Data and calculation conventions
Return and drawdown statistics cover completed cycles. Censored positions are listed separately in the counts and trade tables. Average and worst per-trade drawdowns are calculated for each holding interval. The models use the same score dates, while their actual holding intervals are determined by their own signals.
Returns use closing prices and exclude dividend reinvestment, commissions, and slippage; 1321 is the Nikkei ETF proxy. Weekly data enters using a proxy for publication timing, and historical revision snapshots are incomplete. This article is exploratory historical research.