Markets are not static.
They look like charts on a screen, but underneath those charts sits a living system of participants, capital, rules, data, execution speed, liquidity, and behaviour. Every trader observes the market, forms a view, takes a position, receives feedback through P&L, and then changes behaviour.
That loop never stops.
This is the simplest way to understand why an algo can perform beautifully in one market environment and then behave differently when the environment changes. An algo is not trading a fixed machine. It is trading a market that is continuously learning.
At Stratzy, we think of markets as information-processing systems. Capital moves towards patterns. Once enough capital understands a pattern, the pattern itself changes.
That is alpha decay.
- What is alpha decay?
Alpha is the excess return a strategy earns because it sees, interprets, or acts on something better than the rest of the market.
Decay happens when that advantage reduces.
A simple historical example is RSI. The Relative Strength Index was introduced by J. Welles Wilder Jr. in 1978 in his book New Concepts in Technical Trading Systems. At that time, simply being able to calculate such indicators systematically and act on them had value because computers were not everywhere.
Today, RSI is available to almost every retail trader on every charting platform.
The indicator did not disappear. But the easy edge from just calculating it faster or mechanically trading it reduced because the market learned it.
This is the key idea:
An alpha is not a permanent object. It is a temporary advantage inside a changing system.
- Why does alpha decay?
There are six broad reasons.
A. Logic leak
If the same logic is used by too many participants, everyone starts fighting for the same opportunity.
Suppose a strategy sells options every time implied volatility rises beyond a certain threshold. If only a few participants do this, the edge can remain meaningful. But if thousands of participants crowd into the same signal, the entry gets worse, fills get worse, and the opportunity compresses.
The trade may still be logically correct. It may simply no longer be as profitable because too many people reached the same conclusion.
B. Market microstructure change
This is one of the most misunderstood reasons for decay.
Market microstructure means the actual plumbing of the market:
- when orders can be placed
- how closing prices are calculated
- when liquidity appears
- how participants hedge
- what order types are allowed
- how risk controls work
- how market makers price uncertainty
The new Closing Auction Session, or CAS, is a good example.
NSE has introduced CAS in the equity cash segment for eligible stocks. In Phase 1, it applies to stocks in the cash segment on which derivative contracts are available. The CAS window runs from 3:15 pm to 3:35 pm. Equity derivatives trading runs till 3:40 pm.
This matters because the last part of the day is not just “more time”. It is a different type of time.
Earlier, closing price discovery happened through continuous trading behaviour near the close. With CAS, eligible cash stocks move into an auction-style closing process where orders are collected and matched at an equilibrium price. That changes how the closing print forms. When the closing print formation changes, hedging behaviour, option pricing behaviour, and late-day liquidity behaviour can also change.
So if an algo was built around old end-of-day behaviour, it may face a different market after CAS.
This does not mean the algo was bad. It means the market rules changed, and when rules change, participant behaviour changes.
Visual idea: What changed near the close
| Time window | What happens now | Why it matters for algos |
|---|---|---|
| 3:00 - 3:15 pm | Reference price is formed using VWAP for eligible cash stocks | This becomes the anchor for the auction |
| 3:15 - 3:20 pm | Transition from continuous trading to CAS | Old end-of-day behaviour starts changing here |
| 3:20 - 3:30 pm | Orders are collected in CAS | Liquidity is being gathered, not continuously discovered in the same way |
| 3:30 - 3:35 pm | Orders are matched and closing price is discovered | Closing print comes from auction equilibrium |
| Till 3:40 pm | Equity derivatives continue trading | Options can still reprice around the final cash-market close |
The practical takeaway is simple: if an algo had an edge in the old closing behaviour, we should re-check how that edge behaves in the new closing behaviour.
C. Data degradation
Some alphas come from unique data.
If a data source is private, clean, and early, it can give an edge. But once the same data becomes widely available, cleaned by vendors, and plugged into many models, it stops being rare.
The market has absorbed it.
The same thing happens with price data, news data, option-chain data, alternative data, and even behavioural signals.
D. Overfitting
Overfitting happens when a strategy is tuned too tightly to the past.
For example, if a rule works only because a parameter was adjusted to match a specific historical period, it may fail when the next period is slightly different. The backtest looks great, but the live market was never actually understood.
This is why testing matters.
At Stratzy, we do not look only at headline returns. We evaluate robustness, drawdowns, live behaviour, slippage assumptions, regime performance, risk limits, and capacity before an algo is made available. A good algo should not be a curve-fitted memory of the past. It should be a structure that has survived different kinds of market behaviour.
E. Execution and latency bleed
Sometimes the signal is still valid mathematically, but the execution layer leaks the edge.
In fast-moving option markets, a delay of even a few milliseconds can matter when prices are shifting quickly. If the algo sees a signal but order placement, broker routing, or confirmation is slow, the trade may execute at a worse price than expected.
This is why infrastructure is part of alpha.
Research creates the signal. Execution preserves it.
At Stratzy, we are continuously improving our architecture, routing, and execution systems. Our roadmap is to make execution scalable for thousands of users with sub-100 ms level internal processing after upcoming infrastructure upgrades. The goal is simple: if an alpha exists, the system should not lose it because of avoidable execution drag.
F. Crowding and capacity
Every strategy has capacity.
A strategy that works well with Rs. 10 crore may not work the same way with Rs. 1,000 crore if liquidity is limited. As more capital follows the same signal, the act of entering and exiting the trade starts moving the opportunity itself.
This is why capacity testing matters.
Stratzy currently focuses heavily on highly liquid option structures on instruments like NIFTY, where market depth is significantly better than most instruments. This helps reduce capacity pressure in the early scaling phase. But no strategy has infinite capacity, which is why user slots, allocation limits, and algo-level monitoring are important.
3. What can we control, and what can we not control?
This is the honest framework.
Some things are controllable.
We can control research quality. We can control testing discipline. We can control execution infrastructure. We can control position sizing. We can control capacity checks. We can control when to pause, reduce, or retire an algo.
Some things are not controllable.
We cannot control SEBI or exchange rule changes. We cannot control sudden macro events. We cannot control a global headline moving the market. We cannot control whether a market regime shifts from trending to mean-reverting, or from calm to jumpy.
But we can control how quickly we adapt.
That is the real game.
Alpha research is not a one-time product. It is a loop:
- discover alpha
- test alpha
- deploy carefully
- monitor decay
- manage capacity
- reduce or pause when behaviour changes
- introduce newer alphas for the new environment
The market keeps learning. So the alpha pool also has to keep evolving.
4. Where does correlation fit in?
Many users ask us: “If two algos are different, why are they sometimes correlated?”
The answer is that correlation is not decided by the name of the algo. It is decided by how the algo makes and loses money.
At Stratzy, we prefer looking at daily P&L correlation.
In simple words:
If two algos tend to make money and lose money on the same days, they are more correlated.
They may have different names. They may use different strikes. They may enter at different times. But if both depend on the same market behaviour, their P&L can still move together.
5. A simple correlation framework
When combining algos, think across three layers.
A. Instrument diversification
This means using different underlying instruments.
For example:
- NIFTY-based algos
- SENSEX-based algos
- BANKNIFTY-based algos
- stock-option based algos
Different instruments often respond differently to flows, news, liquidity, and volatility. This can reduce correlation.
B. Structure diversification
This means using different trade structures.
For example:
- credit spreads
- debit spreads
- strangles
- iron condors
- directional option structures
- intraday scalping structures
Two credit-spread algos may look different because their strikes, entry rules, or timings differ. But structurally, they may still be exposed to similar risks: direction, volatility, gap movement, and late-day option repricing.
Combining a credit-spread algo with a strangle algo or a directional algo may create better diversification than combining only multiple versions of the same structure.
C. Timing and regime diversification
This means using algos that operate in different market windows or different regimes.
For example:
- morning volatility algos
- mid-day mean reversion algos
- expiry-day algos
- non-expiry algos
- event-sensitive algos
- end-of-day algos
Timing matters because market behaviour changes through the day. Liquidity, hedging, gamma exposure, volatility decay, and institutional flows are not the same at 9:30 am and 3:20 pm.
This becomes especially important after market-structure changes like CAS.
Visual idea: The correlation ladder
| Portfolio mix | Expected correlation | Why |
|---|---|---|
| Same instrument + same structure + same timing | Higher | The algos are exposed to similar market moves |
| Same instrument + same structure + different timing | Medium-high | Timing differs, but the core payoff is still similar |
| Same instrument + different structure + different timing | Medium | The underlying is same, but payoff behaviour changes |
| Different instrument + different structure + different timing | Lower | More independent sources of return and risk |
This is why diversification is not just adding more algos. It is adding different return engines.
6. Why “different algos” does not always mean “diversified”
This is a subtle but important point.
Suppose you use five different credit-spread algos. They may enter at different times and choose different strikes. But in a strong trending move, all five may end up taking similar directional exposure or facing similar option-pricing stress.
So diversification is not about counting algos.
It is about counting independent sources of risk and return.
A better question is not:
“How many algos do I have?”
A better question is:
“How many different ways can my portfolio make money, and how many different ways can it lose money?”
That is the core of portfolio thinking.
7. How do we tackle alpha decay and correlation at Stratzy?
Our approach has four parts.
A. Keep expanding the alpha pool
Stratzy currently has 200+ algos in its research and deployment universe, and we are working towards expanding this to 1000+ researched algos.
The goal is not to randomly create more algos. The goal is to create a larger menu of genuinely different alpha sources across instruments, structures, timings, and market regimes.
When one pocket of alpha decays or becomes temporarily unstable, the platform should have other researched alternatives.
B. Monitor live behaviour, not just backtests
Backtests are useful, but live markets are the real examination.
We track whether live behaviour remains consistent with expected behaviour. If the market changes, we do not want to defend an old model emotionally. We want to observe, reduce, pause, improve, or replace.
C. Manage capacity
If too much capital enters the same algo, performance can degrade. This is why capacity testing, slot limits, and allocation controls matter.
Alpha should be scaled carefully, not carelessly.
D. Build adaptive allocation
This is where our upcoming AlgoBoxes become important.
Instead of users manually deciding every allocation across multiple algos, AlgoBoxes are designed to manage allocation across algos and adapt as market conditions change.
The long-term direction is clear: users should not need to chase one perfect algo. They should be able to access a managed basket of researched alphas, diversified across structures and market regimes.
8. The final mental model
Think of the market as a learning machine.
Every participant watches the same prices, reacts to the same events, and learns from the same P&L feedback. When a pattern becomes visible, capital moves towards it. When capital moves towards it, the pattern changes.
So the question is not:
“Will this algo work forever?”
The better questions are:
- What is the source of this alpha?
- What can cause this alpha to decay?
- Is the strategy robust across market regimes?
- Is the strategy capacity-constrained?
- How correlated is it with my other algos?
- What happens if the market microstructure changes?
- Does my portfolio have multiple independent return sources?
That is how professional investors think.
No algo is permanent. But a disciplined alpha engine can keep adapting.
That is what we are building at Stratzy.