Private‑Credit Impulse

Abstract

The Private‑Credit Impulse signal quantifies acceleration or deceleration in aggregate private‑sector credit growth. Version 1.1 blends a short‑horizon momentum measure (3‑month annualised growth) with a slower trend measure (year‑over‑year growth) for total bank credit, business lending, and consumer credit. Each growth series is robust‑standardised and averaged to create a composite that identifies Accelerating, Stable, and Decelerating credit regimes, with simple confirmation rules to reduce whipsaws.

1. Data

2. Growth‑Rate Transformation

Credit aggregates are level series. To separate trend from momentum, v1.1 computes two growth transforms for each component:

For legacy continuity, you may also compute the prior six‑month annualised growth:

Revision note: weekly bank credit series are frequently revised. Short-horizon growth (3m/6m annualised) should be interpreted with a revision-risk discount, especially at the sample endpoint.

3. Normalisation

4. Composite Construction

1. Compute g^{yoy} and g^{3m} growth for each input series.
2. Macro adjustment (recommended): deflate nominal credit growth by inflation and benchmark against GDP growth:
   • g^{real} \approx g^{nominal} - \pi (using CPI or PCE inflation \pi).
   • g^{excess} \approx g^{nominal} - (\pi + g^{real\ GDP}) to approximate credit growth above “nominal trend”.
   This reduces false “acceleration” signals in high-inflation or strong-nominal-GDP environments.
3. Robust-standardise each growth series with a rolling median/MAD z‑score.
4. Build two composites:
   • Trend composite: average of component z‑scores on g^{yoy}.
   • Momentum composite: average of component z‑scores on g^{3m}.
5. Combine trend and momentum:
   C_t = w_m\,C^{mom}_t + (1-w_m)\,C^{trend}_t

   Default w_m=0.6 (momentum-weighted) to keep the signal responsive while anchoring to the slower YoY trend.

5. Regime Mapping

Regimes map the composite C_t into discrete states. v1.1 introduces a small transition band and recommends macro-adjusted calibration.

• Default threshold: h=0.75 (heuristic).
• Transition / hysteresis: require two consecutive monthly closes beyond ±h to confirm a new regime, and treat |C_t|\in[0.60,0.75] as “transition risk” rather than a clean flip.
• Inflation/GDP-aware calibration: if using nominal growth without macro adjustment, consider widening h during high inflation or unusually strong nominal GDP growth to avoid overstating “credit acceleration”.

6. Implementation (Python)

import pandas as pd
import numpy as np

def robust_z(s, win=48, min_win=18):
    x = pd.to_numeric(s, errors="coerce").astype(float)
    w = max(min_win, min(win, x.dropna().size))
    med = x.rolling(w, min_periods=min_win).median()
    mad = (x - med).abs().rolling(w, min_periods=min_win).median()
    return (x - med) / (1.4826 * mad.replace(0, np.nan))

def yoy(series):            # trend
    return (series / series.shift(12)) - 1.0

def ann3(series):           # momentum (3m annualised)
    return (series / series.shift(3))**4 - 1.0

d = df_priv_credit.copy()

# Optionally join CPI/PCE inflation and real GDP growth to compute real/excess measures.
# Example (approx): g_real = g_nominal - inflation; g_excess = g_nominal - (inflation + real_gdp)

components = ["TOTBKCR","BUSLOANS","TOTALSL"]

# growth transforms
for col in components:
    d[f"{col}_yoy"]  = yoy(d[col])
    d[f"{col}_ann3"] = ann3(d[col])

# z-score each transform
z_trend, z_mom = [], []
for col in components:
    zt = robust_z(d[f"{col}_yoy"])
    zm = robust_z(d[f"{col}_ann3"])
    d[f"Z_{col}_trend"] = zt
    d[f"Z_{col}_mom"]   = zm
    z_trend.append(f"Z_{col}_trend")
    z_mom.append(f"Z_{col}_mom")

d["C_trend"] = d[z_trend].mean(axis=1)
d["C_mom"]   = d[z_mom].mean(axis=1)

w_m = 0.6
d["Private_Credit_Impulse"] = w_m*d["C_mom"] + (1-w_m)*d["C_trend"]

h = 0.75

def regime(series, hi=h, confirm=2):
    # two-close confirmation to reduce whipsaws
    out = pd.Series(index=series.index, dtype="object")
    state = "Stable"
    count = 0
    for i, v in enumerate(series.values):
        if pd.isna(v):
            out.iat[i] = np.nan
            continue
        target = "Accelerating" if v > hi else ("Decelerating" if v < -hi else "Stable")
        if target == state:
            count = 0
        else:
            count = count + 1 if target != "Stable" else 0
            if target == "Stable":
                state, count = "Stable", 0
            elif count >= confirm:
                state, count = target, 0
        out.iat[i] = state
    return out

d["Credit_Impulse_Regime"] = regime(d["Private_Credit_Impulse"])

df_sig_priv_credit = d
display(df_sig_priv_credit.tail())

7. Interpretation

8. Evidence base & model interpretation

9. Limitations

• Weekly bank credit series are subject to revisions; near-term regime classifications can change as reporting is updated. Consider real-time (ALFRED) evaluation for backtests.
• Loan aggregates can shift due to accounting, securitisation, and reclassifications; these can create artificial swings unrelated to credit supply/demand.
• Inflation and nominal GDP regimes affect nominal credit growth. If you do not apply macro adjustments, static thresholds can misclassify high-inflation periods as “accelerating.”
• Fixed component weights and heuristic thresholds are simplifications; periodic recalibration and publication of component contributions improve robustness and transparency.