Table 7 puts the paper's strongest result on public long-term debt, while the abstract assigns it to the wrong maturity bucket. The coefficients are -2.713 for positive shocks and -10.333 for negative shocks, with Wald p<0.001. Public short-term debt is insignificant in both directions and symmetric. A reader relying on the abstract would leave with the central result reversed.

Bulut studies whether external debt affects Turkish bank equity differently by maturity and borrower type. XBANK, Borsa Istanbul's banking sub-index, is the dependent variable. NARDL splits each regressor into the running sum of its rises and the running sum of its falls, allowing the two halves to have different long-run coefficients.

Four models use quarterly data from 1997Q1 to 2024Q4, giving 112 observations, with USD/TRY, VIX and WTI as controls. Bulut reports significant long-run asymmetry for private short-term debt (Wald p=0.031) and public long-term debt (Wald p<0.001).

The funding channel and its measurement

A Turkish bank funded partly offshore faces rollover risk and currency mismatch when external debt is short-dated. Growth in the sovereign's long-dated debt stock creates another route, as the country risk premium reprices. Funding costs transmit both effects into credit supply and then bank equity. XBANK is a reasonable barometer for that chain.

The sample again covers 1997Q1 to 2024Q4, with 112 quarterly observations. Bulut divides the external debt stock four ways: public short (KK), public long (KU), private short (OK) and private long (OU). The controls are USD/TRY, VIX and WTI. In the data section, the last is labelled "Brent ham petrol (WTI)", although those are two different crude benchmarks.

All series are logged and left seasonally unadjusted because only lnOK showed weak seasonality (F=3.266, p=0.024). ADF, KPSS and Zivot-Andrews are used for stationarity checks. A BDS test on residuals from a linear model supplies the case for nonlinearity (z=11.601 at m=2, p=0.000). Cointegration is assessed through the Pesaran et al. (2001) bounds test. The asymmetry comes from NARDL partial sums following Shin et al. (2014), with Wald tests applied afterward. Newey-West standard errors are used throughout because three of four models fail Breusch-Pagan-Godfrey (lnKK p=0.023, lnOK p=0.033, lnOU p=0.006).

The study reports an in-sample relationship between debt stocks and an index level. There is no portfolio or return series, and no costs are charged.

What does Table 7 actually show?

For public short-term debt, the long-run positive component is 0.400 (p=0.177) and the negative component is 0.661 (p=0.101). The Wald result is p=0.297. Those estimates provide no significant directional effect or asymmetry.

Public long-term debt carries the result: a positive component of -2.713 (t=-3.74), a negative component of -10.333 (t=-4.83), and Wald p<0.001. The conclusion section describes this correctly, saying that public short-term debt has a limited effect while public long-term debt produces strong negative asymmetric effects. The paper therefore contradicts its own abstract in Turkish and in English. The error appears again when the conclusion says short-term public and private debt generate different asymmetric effects, even though the public leg is the insignificant result.

The second abstract claim fares somewhat better. Positive shocks to private long-term debt have a coefficient of 1.128 (t=2.02, p=0.045), supporting the index. Yet Bulut reports the Wald test for that pair as insignificant, and the text says so directly. The abstract illustrates its asymmetry claim with two results, while the paper's own Wald test classifies one of those pairs as symmetric. It is selling a different paper.

Table 7 contains another issue that an editor should resolve first. Private short-term debt has long-run coefficients of 1.128 and 1.899. Private long-term debt also has coefficients of 1.128 and 1.899. They are identical to three decimals, though their t-statistics differ (2.82 versus 2.02, 3.51 versus 3.29). I read this as a transcription error; two independent regressions landing on the same pair of estimates seems less likely.

Four debt series, four regressions

The abstract says the sector and maturity splits are treated "simultaneously". The tables show separate estimation. The bounds test table lists seven univariate models, one for each regressor: lnXBANK ~ lnKK, lnXBANK ~ lnKU, and so on. The diagnostics table labels Model 1 through Model 4, one per debt series.

Public and private external debt, at short and long maturities, share trends and break episodes. Zivot-Andrews dates lnKU at 2004Q4, lnOU at 2008Q2 and lnXBANK at 2020Q3. Estimating each debt series one at a time lets its coefficient absorb the others. In that setting, I read the -10.333 estimate on negative public long-term shocks as the product of two I(1) series left alone together. ADF, KPSS and Zivot-Andrews all classify the debt stocks and XBANK as I(1). The error-correction coefficient is never reported, and the sample contains 112 quarters.

The bounds decisions use a 5% upper bound of 4.85. Five of the seven models clear it comfortably: lnKK 7.054, lnKU 7.419, lnOK 5.472, USD/TRY 5.088 and VIX 5.801. The other two create problems. OU records 4.596, between the bounds, yet is classified as cointegrated. WTI records 3.592, below the 3.79 lower bound, yet is classified as borderline.

Other table entries also conflict. The ADF decision column marks lnVIX as I(1), although the statistics in the same row are -4.297 (p=0.003) and -4.134 (p=0.001). The surrounding text correctly describes VIX as I(0). Public long-term debt has a third quartile of 1,133.20 and a minimum of 38,552 in the descriptive table. For lnKK, the seasonality table reports an F-statistic of 122 and a p-value of 946. We did not find an error-correction coefficient or its t-statistic anywhere in the paper, despite its role as the standard validity check for a NARDL long run.

A level problem remains underneath these results. XBANK is a nominal lira index ranging from 16.50 to 14,533.69. Over the same sample, USD/TRY ranges from 0.12 to 34.76, while the debt stocks are denominated in US dollars. Their estimated long-run relationship risks giving an economic label to a shared inflation and currency trend.

Why we left it alone

We could not reproduce any of these results. The dependent variable is Turkish bank equity, and the regressors are TCMB EVDS external debt stocks divided by sector and original maturity. Neither series exists in our data. Substituting US financials and US macro series would examine a different mechanism, so it would not constitute a replication.

One joint NARDL would change my view if it included all four debt components and the controls in a single specification, using real rather than nominal series. It should report the error-correction term with its t-statistic, along with dynamic multiplier paths showing how long a public long-term shock takes to reach XBANK. If the public long-term asymmetry survives that setup, the paper has a country-risk signal worth watching. For now, its strongest finding is the one the abstract leaves out.